In the din of heated public debates about artificial intelligence, from speculative doomsday predictions to starry-eyed visions of an AI utopia, people on both sides of the argument tend to position society as an object of this “inevitable” transformation: We are all passive witnesses to this onward march of technology, which we endow with almost divine agency.

But as we’ve argued carefully throughout this report, this is a fight about power, not progress. We need not reconcile ourselves to passive alarm, nor engage in heedless celebration of speculative futures: Instead, the path ahead requires reclaiming agency over the trajectory for AI. In order to build a movement toward a just and democratic society, we must contest how AI is used to avert it. Only when we do so can we seed a new path defined by autonomy, dignity, respect, and justice for all of us. 

Here, we offer a high-level road map of strategic levers and case studies to illuminate the opportunity before us. Some of these speak more directly to advocacy campaigns and narrative strategy; others offer regulatory directions for reining in and rebalancing unaccountable power and rebalancing power:

Let’s examine each of these in turn.

Target How the AI Industry Works Against the Interests of Everyday People

After decades of mounting inequity, the foundations of a good life have been priced out of reach for most people to such an extent that, in the aftermath of the 2024 election, there’s growing consensus across the ideological spectrum that focusing on the material conditions and economic interests of working people is key to building political power.1See Elizabeth Wilkins, “Restoring Economic Democracy: Progressive Ideas for Stability and Prosperity,” Roosevelt Institute, April 29, 2025, (<)a href='https://rooseveltinstitute.org/publications/restoring-economic-democracy/'(>)https://rooseveltinstitute.org/publications/restoring-economic-democracy(<)/a(>); Ezra Klein, “Would Bernie Have Won?” (<)em(>)Ezra Klein Show,(<)/em(>) November 26, 2024, (<)a href='https://www.nytimes.com/2024/11/26/opinion/ezra-klein-podcast-faiz-shakir.html'(>)https://www.nytimes.com/2024/11/26/opinion/ezra-klein-podcast-faiz-shakir.html(<)/a(>); Pippa Norris et al., “Trump, Brexit, and the Rise of Populism: Economic Have-Nots and Cultural Backlash,” unpublished research project, HKS Faculty Research Working Paper Series RWP16-026, Harvard Kennedy School, 2016; and George Hawley, “The Political Economy of Right-Wing Populism in the United States,” (<)em(>)Promarket(<)/em(>), May 24, 2024, (<)a href='https://www.promarket.org/author/george_hawley/'(>)https://www.promarket.org/author/george_hawley(<)/a(>). While the Trump campaign was successful in moving a pro-working-class message, this might in fact be the administration’s core vulnerability2Patrick Wintour, “Did Trump’s Tariffs Kill Economic Populism?” (<)em(>)Guardian(<)/em(>), April 12, 2025, (<)a href='https://www.theguardian.com/world/2025/apr/12/did-trump-tariffs-economic-populism-globalisation'(>)https://www.theguardian.com/world/2025/apr/12/did-trump-tariffs-economic-populism-globalisation(<)/a(>); Alexander R. Ross, “Steve Bannon and Elon Musk Are Battling for the Soul of Trumpism,” (<)em(>)New Lines Magazine(<)/em(>), April 15, 2025, (<)a href='https://newlinesmag.com/argument/steve-bannon-and-elon-musk-are-battling-for-the-soul-of-trumpism/'(>)https://newlinesmag.com/argument/steve-bannon-and-elon-musk-are-battling-for-the-soul-of-trumpism(<)/a(>)., given that much of the policy orientation of the administration so far has been successfully commanding loyalty from tech and financial elites. There is momentum and urgency around finding ways to drive a wedge through this hypocrisy. The road map we lay out in this section explores ways to ensure that the AI fight revolves around the tech industry’s amassing of power that profoundly affects the day-to-day lives of people—not around progress. 

That the AI industry is fundamentally incentivized to hurt the interests of working people and families appears not to be widely understood yet—partly because the most high-profile narratives around AI risk relate to issues of technically framed bias or existential risks, which are often disconnected from people’s material realities. And the fact that AI expertise largely circulates in elite and urban corridors only exacerbates the problem. As we’ve shown throughout this report, investment and interest in AI is not about supporting progress for the many, but about hoarding power for the few, and doing so by any means available: asymmetric control over information; gatekeeping of infrastructure; hollowing out agency for citizens, workers, and consumers; or supporting political allies, financially or otherwise, who side with Big Tech. 

For those of us whose research and advocacy is centered around AI, this means that we need not only to make AI-related issues more relevant to movements fighting for economic populism and against tech oligarchy; we also need to better target the AI industry as a key actor working against the interests of the working public. The writing is on the wall: We need to prioritize policy issues that are rooted in people’s lived experiences with AI, particularly those that hit at their most urgent material conditions. This involves giving the broader public a way to “see” AI systems as the invisible infrastructures that mediate their lives—often for the worse—and helping them connect the dots between the harms of AI and the unchecked power of Big Tech. Fortunately, there are existing windows of opportunity for doing just this:

1. DOGE Pushback

The evisceration of government agencies by DOGE brings home the harms that tech oligarchs can enact on the public. This makes it an important front for resistance to AI. There is widespread discontent with the workings of DOGE teams even among those who advocate for greater government efficiency;3Ezra Klein, “In This House, We’re Angry When Government Fails,” (<)em(>)Ezra Klein Show(<)/em(>), November 22, 2024, (<)a href='https://www.nytimes.com/2024/11/22/opinion/ezra-klein-podcast-jennifer-pahlka-steven-teles.html'(>)https://www.nytimes.com/2024/11/22/opinion/ezra-klein-podcast-jennifer-pahlka-steven-teles.html(<)/a(>). and reporting thoroughly documents the failure of DOGE to achieve its stated goals.4 David A. Fahrenthold and Jeremy Singer-Vine, “DOGE is Far Short of Its Goal, and Still Overstating Its Progress, (<)em(>)New York Times(<)/em(>), April 13, 2025, (<)a href='https://www.nytimes.com/2025/04/13/us/politics/doge-contracts-savings.html?smid=nytcore-ios-share&referringSource=articleShare'(>)https://www.nytimes.com/2025/04/13/us/politics/doge-contracts-savings.html?smid=nytcore-ios-share&referringSource=articleShare(<)/a(>). While DOGE officials claim that “AI” can be used to identify budget cuts, detect fraud,5Kate Conger, Ryan Mac, and Madeleine Ngo, “Musk Allies Discuss Deploying A.I. to Find Budget Savings,” (<)em(>)New York Times(<)/em(>), February 3, 2025, (<)a href='https://www.nytimes.com/2025/02/03/technology/musk-allies-ai-government.html'(>)https://www.nytimes.com/2025/02/03/technology/musk-allies-ai-government.html(<)/a(>). automate government tasks,6Makena Kelly and Zoe Schiffer, “DOGE Has Deployed Its GSAi Custom Chatbot for 1,500 Federal Workers,” (<)em(>)Wired(<)/em(>), March 7, 2025, (<)a href='https://www.wired.com/story/gsai-chatbot-1500-federal-workers/'(>)https://www.wired.com/story/gsai-chatbot-1500-federal-workers(<)/a(>). and determine whether someone’s job is “mission critical,”7Courtney Kube et al., “DOGE Will Use AI to Assess the Responses of Federal Workers Who Were Told to Justify Their Jobs via Email,” NBC News, February 25, 2025, (<)a href='https://www.nbcnews.com/politics/doge/federal-workers-agencies-push-back-elon-musks-email-ultimatum-rcna193439'(>)https://www.nbcnews.com/politics/doge/federal-workers-agencies-push-back-elon-musks-email-ultimatum-rcna193439(<)/a(>). reporting has dismantled over and over again that both these teams and the technology they are using are not up to the task.8Aatish Bhatia et al., “DOGE’s Only Public Ledger is Riddled with Mistakes,” (<)em(>)New York Times(<)/em(>), February 28, 2025, (<)a href='https://www.nytimes.com/2025/02/21/upshot/doge-musk-trump-errors.html'(>)https://www.nytimes.com/2025/02/21/upshot/doge-musk-trump-errors.html(<)/a(>); Fahrenthold and Singer-Vine, “DOGE Far Short of Its Goal.” And while it is still unclear exactly what DOGE employees are doing with the data they’ve gained access to, it is very clear that much of this data is highly lucrative and could be used to further Musk’s business interests—including where it is stored, and whether it is by being used to train AI models for Musk’s own companies.9Donald S. Beyer Jr. et al., to Russell Vought, February 12, 2025, (<)a href='https://beyer.house.gov/uploadedfiles/letter_from_congress_to_omb_director_on_restoring_public_access_to_federal_data.pdf'(>)https://beyer.house.gov/uploadedfiles/letter_from_congress_to_omb_director_on_restoring_public_access_to_federal_data.pdf(<)/a(>). 

The outrage over DOGE offers an entry point not only to draw attention to the harm government agencies are creating with their use of AI, but also to make clear that DOGE’s work has never truly been about making our services more “efficient”—the goal has always been to dismantle government services and centralize power (see Chapter 3).10Robert Weissman, “DOGE Delusions, A Real-World Plan to Crack Down on Corporate Handouts, Tax the Rich and Invest for the Future,” Public Citizen, January 15, 2025, (<)a href='https://www.citizen.org/article/doge-delusions/'(>)https://www.citizen.org/article/doge-delusions(<)/a(>). Building campaigns that connect the pushback against DOGE to the broader pattern of hollowing out social services in the name of AI-enabled “efficiency”’—from child services, to provision of housing, to access to healthcare—with partners like American Federal Government Employees, Federal Unionists Network, and Federal Workers Against DOGE is a crucial starting point. Many of these partners have already made significant in-roads in helping the public understand and mobilize around DOGE’s harm to workers; broadening the discussion to include how Musk is using AI to help the Trump Administration enact an austerity agenda that impoverishes and disenfranchises the working class, communities of color, the disabled, and those in rural communities could move the needle even further.11Kevin De Liban, (<)em(>)Inescapable AI(<)/em(>), TechTonic Justice, November 2024, (<)a href='https://www.techtonicjustice.org/reports/inescapable-ai'(>)https://www.techtonicjustice.org/reports/inescapable-ai(<)/a(>).

2. Data Centers

The unchecked growth of AI infrastructure is proof of how incredibly successful Big Tech has been at making the case that AI—especially large-scale AI—is worthy of exceptional support and investment from public actors. But the reasons why large-scale AI is a worthwhile use of resources, taxpayer dollars, and land are all based on potential economic growth, the projections of which are usually based on unfounded assumptions (See Chapter 2: Heads I Win, Tails You Lose). Meanwhile, the harms are definitive and documented, such as those evidenced by Meta and Blackstone’s data centers in rural Georgia12“I Live 400 Yards From Mark Zuckerberg’s Massive Data Center,” More Perfect Union, March 27, 2025, YouTube video, (<)a href='https://www.youtube.com/watch?v=DGjj7wDYaiI&ab_channel=MorePerfectUnion'(>)https://www.youtube.com/watch?v=DGjj7wDYaiI&ab_channel=MorePerfectUnion(<)/a(>). (see Chapter 1.2: Too Big To Fail and Chapter 1.3: Arms Race 2.0).13OpenAI, “OpenAI’s Infrastructure Blueprint for the US,” November 13, 2024, (<)a href='https://media.datacenterdynamics.com/media/documents/OpenAI_Blueprint-DCD.pdf'(>)https://media.datacenterdynamics.com/media/documents/OpenAI_Blueprint-DCD.pdf(<)/a(>); Abeba Birhane et al., “The Forgotten Margins of AI Ethics,” in (<)em(>)2022 ACM Conference on Fairness, Accountability, and Transparency(<)/em(>) (2022): 948–958, (<)a href='https://dl.acm.org/doi/10.1145/3531146.3533157'(>)https://dl.acm.org/doi/10.1145/3531146.3533157(<)/a(>); and Abeba Birhane, “AI: Potential Benefits, Proven Risks, Abeba Birhane,” Better Ways (Agile Greece Summit), March 12, 2025, YouTube video, (<)a href='https://www.youtube.com/watch?v=P-p_n4XU0Y8&ab_channel=BetterWays%28AgileGreeceSummit%29'(>)https://www.youtube.com/watch?v=P-p_n4XU0Y8&ab_channel=BetterWays%28AgileGreeceSummit%29(<)/a(>). 

The ground is ripe for local mobilization around the material impacts of the AI infrastructure build-out—not only by building, organizing, and campaigning focused on the tech sector but also by forming alliances with the environmental justice movement, which has long and deep experience fighting against policies that result in a physical degradation of natural resources, as well as movements for utility reform and racial justice. Data centers have direct environmental and community health impacts, including (but not limited to) increased greenhouse gas emissions, increased rates of pollution, and the expansion of gas infrastructure and coal plants.14Lois Parshley, “The Hidden Environmental Impact of AI,” (<)em(>)Jacobin(<)/em(>), June 20, 2024, (<)a href='https://jacobin.com/2024/06/ai-data-center-energy-usage-environment'(>)https://jacobin.com/2024/06/ai-data-center-energy-usage-environment(<)/a(>); Stand.earth, “A Growing Climate Concern Around Microsoft’s Expanding Data Center Operations,” December 5, 2024, (<)a href='https://stand.earth/insights/a-growing-climate-concern-around-microsofts-expanding-data-center-operations'(>)https://stand.earth/insights/a-growing-climate-concern-around-microsofts-expanding-data-center-operations(<)/a(>); Yusuf Sar, “The Silent Burden of AI: Unveiling the Hidden Environmental Costs of Data Centers by 2030,” (<)em(>)Forbes(<)/em(>), August 16, 2024, (<)a href='https://www.forbes.com/councils/forbestechcouncil/2024/08/16/the-silent-burden-of-ai-unveiling-the-hidden-environmental-costs-of-data-centers-by-2030'(>)https://www.forbes.com/councils/forbestechcouncil/2024/08/16/the-silent-burden-of-ai-unveiling-the-hidden-environmental-costs-of-data-centers-by-2030(<)/a(>). They also frequently draw on public aquifers for access to the water supply needed to cool their server racks.15Nikitha Sattiraju, “The Secret Cost of Google’s Data Centers: Billions of Gallons of Water to Cool Servers,” (<)em(>)Time(<)/em(>), April 2, 2020, (<)a href='https://time.com/5814276/google-data-centers-water'(>)https://time.com/5814276/google-data-centers-water(<)/a(>); Christopher Tozzi, “How the Groundwater Crisis May Impact Data Centers,” Data Center Knowledge, March 12, 2024, (<)a href='https://www.datacenterknowledge.com/data-center-site-selection/how-the-groundwater-crisis-may-impact-data-centers'(>)https://www.datacenterknowledge.com/data-center-site-selection/how-the-groundwater-crisis-may-impact-data-centers.(<)/a(>) This is particularly concerning given plans to expand data center construction in already water-strained areas of the southwest.16Ireland is another extreme example of this problem. See Heidi Vella, “Ireland’s Data Centre Nightmare – and What Others Can Learn From It,” (<)em(>)Tech Monitor(<)/em(>), February 18, 2025, (<)a href='https://www.techmonitor.ai/hardware/data-centres/irelands-data-centre-nightmare'(>)https://www.techmonitor.ai/hardware/data-centres/irelands-data-centre-nightmare.(<)/a(>) In the Netherlands17“Dutch Call a Halt to New Massive Data Centres, While Rules are Worked Out,” (<)em(>)DutchNews(<)/em(>), February 17. 2022, (<)a href='https://www.dutchnews.nl/2022/02/dutch-call-a-halt-to-new-massive-data-centres-while-rules-are-worked-out'(>)https://www.dutchnews.nl/2022/02/dutch-call-a-halt-to-new-massive-data-centres-while-rules-are-worked-out(<)/a(>); Mark Ballard, “Hyperscale Data Centers Under Fire in Holland,” Data Center Knowledge, February 7, 2022, (<)a href='https://www.datacenterknowledge.com/hyperscalers/hyperscale-data-centers-under-fire-in-holland'(>)https://www.datacenterknowledge.com/hyperscalers/hyperscale-data-centers-under-fire-in-holland.(<)/a(>) and Chile,18“Google Says It Will Rethink Its Plans for a Big Data Center in Chile Over Water Worries,” Associated Press, September 17, 2024, (<)a href='https://apnews.com/article/chile-google-data-center-water-drought-environment-d1c6a7a8e8e6e45257ac84fb750b2162'(>)https://apnews.com/article/chile-google-data-center-water-drought-environment-d1c6a7a8e8e6e45257ac84fb750b2162(<)/a(>). community-level activism led to a pause on data center construction, which suggests that organizing that targets local government can be an effective lever for curtailing the rollout of data centers. These issues—along with other consequences like higher energy costs and labor abuses across the supply chain—are also bipartisan and hit rural voters and Indigenous communities19Chief Sheldon Sunshine to Premier Smith, January 13, 2024, (<)a href='https://www.sturgeonlake.ca/wp-content/uploads/2020/07/13-01-25-Chief-Sunshine-Open-Letter-Premier-Smith-Re-O-Leary.pdf'(>)https://www.sturgeonlake.ca/wp-content/uploads/2020/07/13-01-25-Chief-Sunshine-Open-Letter-Premier-Smith-Re-O-Leary.pdf(<)/a(>); Julia Simon, “Demand for Minerals Sparks Fear of Mining Abuses on Indigenous Peoples’ Lands,”(<)em(>) (<)/em(>)NPR, January 29, 2024, (<)a href='https://www.npr.org/2024/01/29/1226125617/demand-for-minerals-sparks-fear-of-mining-abuses-on-indigenous-peoples-lands'(>)https://www.npr.org/2024/01/29/1226125617/demand-for-minerals-sparks-fear-of-mining-abuses-on-indigenous-peoples-lands.(<)/a(>) particularly hard, providing other critical opportunities for new organizing.

On the flip side, the firms advocating for data center buildout have little proof that data centers actually provide economic benefits to the communities where they are sited. The evidence that industry actors have provided thus far seems speculative at best, often failing to delineate between job projections for full-time roles tied to data center operations and short-term roles in construction and the service sector.20Sarah O’Connor, “Anatomy of a Jobs Promise,” (<)em(>)Financial Times(<)/em(>), January 21, 2025, (<)a href='https://www.ft.com/content/2f25065d-3eeb-49f6-a5eb-8d22ed4697a5'(>)https://www.ft.com/content/2f25065d-3eeb-49f6-a5eb-8d22ed4697a5(<)/a(>). They also ignore the economic impact that higher energy costs and diverted service can have on local businesses.21Saijel Kishan, “‘It’s a Money Loser’: Tax Breaks for Data Centers Are Under Fire.” Bloomberg, May 9, 2024, (<)a href='https://www.bloomberg.com/news/articles/2024-05-09/ai-boom-has-some-states-rethinking-subsidies-for-data-centers'(>)https://www.bloomberg.com/news/articles/2024-05-09/ai-boom-has-some-states-rethinking-subsidies-for-data-centers(<)/a(>); Kasia Tarczynska, “Will Data Center Job Creation Live Up to Hype? I Have Some Concerns.” Good Jobs First, February 12, 2025, (<)a href='https://goodjobsfirst.org/will-data-center-job-creation-live-up-to-hype-i-have-some-concerns/'(>)https://goodjobsfirst.org/will-data-center-job-creation-live-up-to-hype-i-have-some-concerns(<)/a(>); Lulu Ramadan and Sydney Brownstone, “How a Washington Tax Break for Data Centers Snowballed Into One of the State’s Biggest Corporate Giveaways,” (<)em(>)ProPublica(<)/em(>), August 4, 2024, (<)a href='https://www.propublica.org/article/washington-data-centers-tech-jobs-tax-break'(>)https://www.propublica.org/article/washington-data-centers-tech-jobs-tax-break(<)/a(>); Tom Dotan, “The AI Data-Center Boom Is a Job-Creation Bust.” (<)em(>)Wall Street Journal(<)/em(>), February 25, 2025, (<)a href='https://www.wsj.com/tech/ai-data-center-job-creation-48038b67?mod=hp_lead_pos9'(>)https://www.wsj.com/tech/ai-data-center-job-creation-48038b67(<)/a(>) Meanwhile, state and local communities stand to lose billions of dollars in tax revenue as a result of the packages of tax incentives that companies demand before they will agree to build.22Zach Schiller, “Indefensible Tax Breaks for Data Centers Will Cost Ohio,” Policy Matters Ohio, January 7, 2025, (<)a href='https://policymattersohio.org/research/indefensible-tax-breaks-for-data-centers-will-cost-ohio/'(>)https://policymattersohio.org/research/indefensible-tax-breaks-for-data-centers-will-cost-ohio(<)/a(>); “At $1 Billion, Amazon’s Oregon Subsidy is Largest Known in Company’s History,” Good Jobs First, May 19, 2023, (<)a href='https://goodjobsfirst.org/at-1-billion-amazons-oregon-subsidy-is-largest-known-in-history/'(>)https://goodjobsfirst.org/at-1-billion-amazons-oregon-subsidy-is-largest-known-in-history(<)/a(>). This again provides a unique opportunity for broad-based coalitional organizing, as both free-market groups and progressive economic organizations have criticized the disconnect between large corporate tax breaks and the lack of tangible benefits for communities.23Kishan, “It’s a Money Loser”; Jarrett Skorup, “Michigan Has Authorized $4.7 Billion in Taxpayer-Funded Business Subsidies,” Mackinac Center, January 27, 2025, (<)a href='https://www.mackinac.org/pressroom/2025/michigan-has-authorized-4-7-billion-in-taxpayer-funded-business-subsidies'(>)https://www.mackinac.org/pressroom/2025/michigan-has-authorized-4-7-billion-in-taxpayer-funded-business-subsidies(<)/a(>); American Economic Liberties Project, “Tools for Taking on the Corporate Subsidy Machine: Introduction,” September 13, 2022, (<)a href='https://www.economicliberties.us/our-work/subsidy-toolkit-intro/'(>)https://www.economicliberties.us/our-work/subsidy-toolkit-intro(<)/a(>); and Greg LeRoy, (<)em(>)The Great American Jobs Scam: Corporate Tax Dodging and the Myth of Job Creation(<)/em(>) (Berrett Koehler, 2005). Some local organizations have already taken on the mantle: Citizens Action Coalition in Indiana, for example, has called for a moratorium on new hyperscaler data center construction due to the enormous resource constraints and cost burdens that large data centers pose to local communities,24Citizens Action Coalition, “CAC Calls for Data Center Moratorium,” October 15, 2024, (<)a href='https://www.citact.org/news/cac-calls-data-center-moratorium'(>)https://www.citact.org/news/cac-calls-data-center-moratorium(<)/a(>).and Memphis Community Against Pollution in Tennessee has pushed back against the public health and environmental impacts of Elon Musk’s xAI data center buildout near historically Black neighborhoods already facing elevated cancer and asthma risks.25Andrew R. Chow, “Elon Musk’s New AI Data Center Raises Alarms Over Pollution, (<)em(>)Time(<)/em(>), September 17, 2024, (<)a href='https://time.com/7021709/elon-musk-xai-grok-memphis/'(>)https://time.com/7021709/elon-musk-xai-grok-memphis/(<)/a(>).

3. Algorithmic Prices and Wages

Across the economy, from grocery stores to online marketplaces to insurance brokers, the ability of firms to use AI to set prices for customers and calibrate wages for workers using detailed, often intimate, information extracted from us, is deepening inequality. This makes algorithmic prices and wages especially fertile ground for mobilizing around AI’s economic impacts on everyday people.26Veena Dubal, “On Algorithmic Wage Discrimination,”(<)em(>) Columbia Law Review(<)/em(>) 123, no. 7 (2023): 1929–92, (<)a href='https://columbialawreview.org/content/on-algorithmic-wage-discrimination/'(>)https://columbialawreview.org/content/on-algorithmic-wage-discrimination(<)/a(>).

There is growing coordination between labor organizers and anti-monopoly and economic justice groups working to contend with this issue. In February 2025, for example, along with a coalition of organizations and experts, AI Now published a report canvassing the breadth of the use of algorithmic pricing and wage-setting practices and spotlighting the ways in which bright-line rules could prevent harm,27AI Now Institute et al., (<)em(>)AI Now Coauthors Report on Surveillance Prices and Wages(<)/em(>), February 20, 2025, (<)a href='https://ainowinstitute.org/publication/ai-now-coauthors-report-on-surveillance-prices-and-wages'(>)https://ainowinstitute.org/publication/ai-now-coauthors-report-on-surveillance-prices-and-wages(<)/a(>). including an outright ban on individualized surveillance prices and wages, and elimination of loopholes that corporations could exploit to continue that practice.28For example, in cases where states provide narrow exceptions for price or wage discrimination on the basis of surveillance data, businesses must have the burden of clearly establishing that the exception applies. There has already been considerable movement in the states this session to prohibit both algorithmic pricing and wage-setting, including the introduction of bills in California, Colorado, Georgia, and Illinois.29Will Oremus and Lauren Kaori Gurley, “States Eye Bans on ‘Surveillance Pricing’ that Exploits Personal Data,” (<)em(>)Washington Post(<)/em(>), February 20, 2025, (<)a href='https://www.washingtonpost.com/politics/2025/02/20/surveillance-prices-wages-california-ban/'(>)https://www.washingtonpost.com/politics/2025/02/20/surveillance-prices-wages-california-ban(<)/a(>).

Advance Worker Organizing to
Protect the Public and Our Institutions from AI-Enabled Capture

(Guest contribution by Andrea Dehlendorf, Senior Fellow)

Labor campaigns over the past several years have demonstrated that when workers and their unions pay serious attention to how AI is changing the nature of work and engage in serious intervention through collective bargaining, contract enforcement, campaigns, and policy advocacy, they can shape how their employers develop and deploy the technologies. The deeper opportunity for labor, however, lies in the fact that across all sectors of the economy and society, it is people experiencing these technologies at work, firsthand, who best understand their impacts and limitations, and also the conditions under which their potential can be realized. Thus, labor’s power can shape not just whether and how AI is used in the workplace, but can also recalibrate the technology sector’s power overall, ultimately shaping the trajectory of AI toward the public interest and common good. 

Over the past decade, labor campaigns to challenge the use of generative AI in Hollywood, algorithmic management of warehouse workers, and platforms for rideshare drivers have been critical to building public awareness of the impact of AI and data technologies in workplaces. In 2023, the Writers Guild of America and Screen Actors Guild strikes challenged studios aiming to deploy generative AI to replace their creative work without fair compensation.30Molly Kinder, “Hollywood Writers Went on Strike to Protect Their Livelihoods from Generative AI. Their Remarkable Victory Matters for All Workers.” Brookings, April 12, 2024, (<)a href='https://www.brookings.edu/articles/hollywood-writers-went-on-strike-to-protect-their-livelihoods-from-generative-ai-their-remarkable-victory-matters-for-all-workers/'(>)https://www.brookings.edu/articles/hollywood-writers-went-on-strike-to-protect-their-livelihoods-from-generative-ai-their-remarkable-victory-matters-for-all-workers(<)/a(>). These public battles about profit-driven automation risk in creative industries have won contract language that provides some protection for artists and writers, transformed the public conversation around the implications of technology, and shifted popular‬ understanding of the stakes.31Jake Coyle, “In Hollywood Writers’ Battle Against AI, Humans Win (For Now),” Associated Press, September 27, 2023, (<)a href='https://apnews.com/article/hollywood-ai-strike-wga-artificial-intelligence-39ab72582c3a15f77510c9c30a45ffc8'(>)https://apnews.com/article/hollywood-ai-strike-wga-artificial-intelligence-39ab72582c3a15f77510c9c30a45ffc8(<)/a(>). 

Unions, worker centers, and alt-labor groups have also exposed and‬ challenged Amazon’s algorithmic management practices and extensive surveillance. The‬‭ Warehouse Worker Resource Center developed and passed a bill in California that limits the use of‬ technology-enabled productivity quotas—legislation that workers’s organizations are now advancing in other states.32Mishal Khan et al., “The Current Landscape of Tech and Work Policy: A Guide to Key Concepts,” UC Berkeley Labor Center, November 12, 2024, (<)a href='https://laborcenter.berkeley.edu/tech-and-work-policy-guide/'(>)https://laborcenter.berkeley.edu/tech-and-work-policy-guide(<)/a(>); Noam Scheiber, “California Governor Signs Bill that Could Push Amazon to Change Labor Practices,” (<)em(>)New York Times(<)/em(>), October 26, 2021, (<)a href='https://www.nytimes.com/2021/09/22/business/newsom-amazon-labor-bill.html'(>)https://www.nytimes.com/2021/09/22/business/newsom-amazon-labor-bill.html(<)/a(>). Uber and Lyft drivers have also organized across a range of formations (Gig Workers Rising,33Gig Workers Rising, “About Us,” accessed April 17, 2025, (<)a href='https://gigworkersrising.org/get-informed/'(>)https://gigworkersrising.org/get-informed(<)/a(>). Los Deliversistas Unidos,34Workers Justice Project, “Los Deliveristas Unidos,” accessed April 17, 2025, (<)a href='https://www.workersjustice.org/en/ldu'(>)https://www.workersjustice.org/en/ldu(<)/a(>). Rideshare Drivers United,35Drivers United, “We’re Uber & Lyft Drivers Uniting for a Fair, Dignified, and Sustainable Rideshare Industry,” accessed April 17, 2025, (<)a href='https://www.drivers-united.org/'(>)https://www.drivers-united.org(<)/a(>). SEIU, and others) to challenge platform precarity, winning the right to organize and establish labor protections. In Boston, drivers won the right to unionize via a state ballot measure.36Eli Tan,”Ride-Hailing Drivers in Massachusetts Win Right to Unionize,” (<)em(>)New York Times(<)/em(>), November 6, 2024, (<)a href='https://www.nytimes.com/2024/11/06/technology/uber-lyft-drivers-unionize-massachusetts.html'(>)https://www.nytimes.com/2024/11/06/technology/uber-lyft-drivers-unionize-massachusetts.html(<)/a(>). 

As powerful technology corporations take an unprecedented and self-serving role in the federal government, undermining democratic structures, it is clear that the labor movement has an opportunity to lead on behalf of all of society. Building on the momentum of previous wins, there are five strategic paths for the labor movement to flex its power and increase its influence over the trajectory of AI across our society and economy:

Sectoral Analysis and Action for the Common Good

By developing a comprehensive analysis of how AI and digital technologies are advancing in every economic sector, including the impact on both working conditions and the public interest, workers and their unions can develop and implement comprehensive bargaining, organizing, and policy campaign strategies. These strategies will have maximum impact when they address both the impact on the conditions of work and the impact on the public. 

A key example is National Nurses United (NNU), which has pushed back against profit-driven employers seeking to replace skilled professionals with subpar technology that harms patients. In 2024, NNU conducted a survey of its members,37National Nurses United, “National Nurses United Survey Finds A.I. Technology Degrades and Undermines Patient Safety,” May 15, 2024, (<)a href='https://www.nationalnursesunited.org/press/national-nurses-united-survey-finds-ai-technology-undermines-patient-safety'(>)https://www.nationalnursesunited.org/press/national-nurses-united-survey-finds-ai-technology-undermines-patient-safety(<)/a(>). which found that AI undermines nurses’ clinical judgment and threatens patient safety. The group subsequently released a “Nurses’ and Patients’ Bill of Rights,” a set of guiding principles designed to ensure just and safe application of AI in healthcare settings.38National Nurses United, “Nurses and Patients’ Bill of Rights: Guiding Principles for A.I. Justice in Nursing and Health Care,” accessed April 24, 2025, (<)a href='https://www.nationalnursesunited.org/sites/default/files/nnu/documents/0424_NursesPatients-BillOfRights_Principles-AI-Justice_flyer.pdf'(>)https://www.nationalnursesunited.org/sites/default/files/nnu/documents/0424_NursesPatients-BillOfRights_Principles-AI-Justice_flyer.pdf(<)/a(>). NNU affiliates have successfully bargained over AI implementation in contract negotiations; in Los Angeles, nurses secured commitments from University of Southern California hospitals that they won’t replace nurses with AI and that they will center nurses’ judgment when determining patient care, as well as safe and appropriate staffing levels.39Bonnie Castillo, “Humanity is the Heart of Nursing,” National Nurses United, March 29, 2024, (<)a href='https://www.nationalnursesunited.org/article/humanity-is-the-heart-of-nursing'(>)https://www.nationalnursesunited.org/article/humanity-is-the-heart-of-nursing(<)/a(>). At Kaiser Permanente, nurses stopped the rollout of EPIC Acuity, a system that underestimated how sick patients were and how many nurses were needed, and forced Kaiser to establish an oversight committee for the system implementation.40Castillo, “Humanity is the Heart of Nursing.” Beyond the bargaining table, nurses have mobilized their membership out into the streets and to Washington, DC, to protest the hospital industry’s attempts to push scientifically unproven AI instead of investing in safe staffing. 

Similar strategies can be designed across most sectors of the economy, providing a critical check on the deployment of AI for cost cutting and profit maximization.

Challenging AI Austerity in the Public Sector

As Elon Musk and others advance AI-driven austerity in the federal government through DOGE—and Republicans and Democrats mount copycat initiatives across the states—workers and their unions can fight back, defending both workers whose jobs are deeply impacted and those receiving public benefits and services. The Federal Unionist Network‬‭ (FUN‬)‭, an independent network of local unions representing federal employees, is directly challenging DOGE and its driver, Elon Musk. FUN is running a Save Our Services‬ campaign that centers the impact of the service cuts that Musk and the Trump administration are implementing.41Federal Unionists Network, “Let Us Work! Save Our Services!” accessed April 17, 2025, (<)a href='http://www.federalunionists.net/'(>)http://www.federalunionists.net(<)/a(>). The network is building scalable infrastructure for education, support, and action. 

At the state level, Pennsylvania provides a model of regulating AI in the provision of public services, recognizing and protecting the key role state workers play in service and benefit provisions. SEIU Local 668, led by President Steve Catanese, successfully negotiated a partnership with Governor Josh Shapiro establishing a robust co-governance mechanism over AI, including the formation of a worker board that will oversee implementation of generative AI tools; the definition of a public worker as a person and gen AI as a tool, strong human-in-the-loop protections at every step, and monitoring of disparate impacts related to bias and discrimination of protected classes.42Josh Shapiro, “From the Desk of Governor Josh Shapiro,” March 21, 2025, (<)a href='https://www.pa.gov/content/dam/copapwp-pagov/en/governor/documents/2025.3.21%20gov%20shapiro%20letter%20on%20ai%20and%20employees.pdf'(>)https://www.pa.gov/content/dam/copapwp-pagov/en/governor/documents/2025.3.21%20gov%20shapiro%20letter%20on%20ai%20and%20employees.pdf(<)/a(>).

Comprehensive Campaigns
Challenging Big Tech Corporations

Transforming the unprecedented power of the technology sector will require comprehensive campaigns that directly challenge the monopolistic business models and actions of the corporations that dominate and drive the sector, and, increasingly, our government. To meet the challenge, these campaigns must include but not limit themselves to, organizing workers and addressing workplace impacts.

The Athena Coalition, for example, focuses on Amazon’s unaccountable power and impacts across multiple spheres, pushing for regulatory action to break up the business itself.43Athena, “How We Take on the Tech Oligarchy,” February 28, 2025, (<)a href='https://athenaforall.org/news/boycott-alternatives/'(>)https://athenaforall.org/news/boycott-alternatives(<)/a(>). Athena members connect how Amazon develops and uses AI and data surveillance technologies in workplaces to what they sell to police departments, ICE,44#NoTechForIce, “Take Back Tech,” accessed April 17, 2025, (<)a href='https://notechforice.com/'(>)https://notechforice.com(<)/a(>). and authoritarian military states;45No Tech For Apartheid (website), accessed April 17, 2025, (<)a href='https://www.notechforapartheid.com/'(>)https://www.notechforapartheid.com(<)/a(>). they bring public awareness to the impacts of pollution on neighborhoods near Amazon logistic hubs,46“Good Jobs. Clean Air. Justice for Neighbors,” WWRC, accessed April 17, 2025, (<)a href='https://warehouseworkers.org/wwrc_campaigns/building-a-better-san-bernardino/'(>)https://warehouseworkers.org/wwrc_campaigns/building-a-better-san-bernardino(<)/a(>). pressure the company to reduce the climate impact of their business model, and bring together organizers in states to challenge the energy use and environmental impact of their data centers; and they expose how algorithmic systems impact small businesses on the platform, consumers, and workers. Together, coalition members are bringing these strands together in order to reclaim our wealth and power from corporations like Amazon.

Building State and Local Power
to Advance the Public Interest

Workers and their unions can and must collaborate with racial justice, economic justice, and privacy groups to mount a serious, coordinated defense against the tech sector’s aggressive state policy agenda to write the rules of AI regulation and digital labor tech regulation in their interest. Doing so will require deeper engagement with members of unions and community-based organizations to understand and prioritize these interventions. Beyond defense, there is an opportunity to develop a proactive agenda at the state and local level that draws bright lines around the use of surveillance, algorithmic management, and automation technologies in criminal justice, workforce management, and public service delivery; and also addresses the unsustainable climate impacts of data centers and the extreme market concentration of the industry. For example, in California, there is a strong set of labor and public-interest focused policy proposals that can be used as models in other states. Also, engagement is moving with members of unions and aligned privacy, tech justice, racial justice, and economic justice organizations and social justice leaders to understand how their core issues are impacted by AI; develop strategies to organize and publicly campaign on the issues; and identify where they can best engage on state and local policy. The UC Berkeley Labor Center’s Tech and Work Policy Guide is a key resource that can be used across states47Mishal Khan et al., “The Current Landscape of Tech and Work Policy: A Guide to Key Concepts,” UC Berkeley Labor Center, November 12, 2024, (<)a href='https://laborcenter.berkeley.edu/tech-and-work-policy-guide/'(>)https://laborcenter.berkeley.edu/tech-and-work-policy-guide(<)/a(>). and Local Progress’s recent report written in collaboration with AI Now provides an actionable roadmap for local leaders to protect communities from the harms of AI.48Hillary Ronen, Local Leadership in the Era of Artificial Intelligence And the Tech Oligarchy, Local Progress and AI Now Institute, May 2025, (<)a href='https://localprog.org/43FPlS6'(>)https://localprog.org/43FPlS6(<)/a(>).

Organizing Across the AI Value and Supply Chain

A global movement to organize across the AI value and supply chain—from US-based white-collar tech workers to data labelers in the Global South, would provide a powerful check on the technology sector. Such a movement could reorient the trajectory of AI toward the public interest and counter the sector’s authoritarian turn. However, union density in the technology sector, globally and in the US, is negligible. There have been no serious attempts to organize the sector at scale. Existing efforts are often underresourced, while others lack a global orientation and solidarity, limit their focus to economic issues, or are not challenging concentrated market power.

However, several promising initiatives are gaining traction, including Amazon Employees for Climate Justice,49Lauren Rosenblatt, “5 Years into Amazon’s Climate Pledge, Workers Challenge its Progress,” (<)em(>)Seattle Times(<)/em(>), July 10, 2024, (<)a href='https://www.seattletimes.com/business/amazon/5-years-into-amazons-climate-pledge-workers-challenge-its-progress/'(>)https://www.seattletimes.com/business/amazon/5-years-into-amazons-climate-pledge-workers-challenge-its-progress(<)/a(>). which is organizing white-collar workers to challenge Amazon on its climate impact; the African Content Moderators Union (ACMU), other locally led associations,50Mohammad Amir Anwar, “Africa’s Data Workers Are Being Exploited by Foreign Tech Firms – 4 Ways to Protect Them,” (<)em(>)Conversation(<)/em(>), March 31, 2025, (<)a href='https://theconversation.com/africas-data-workers-are-being-exploited-by-foreign-tech-firms-4-ways-to-protect-them-252957'(>)https://theconversation.com/africas-data-workers-are-being-exploited-by-foreign-tech-firms-4-ways-to-protect-them-252957(<)/a(>). and UNI Global,51“African Tech Workers Rising,” UNI Global Union, accessed April 17, 2025, (<)a href='https://uniglobalunion.org/about/sectors/ict-related-services/african-tech-workers-rising/'(>)https://uniglobalunion.org/about/sectors/ict-related-services/african-tech-workers-rising(<)/a(>). which are organizing data labelers in Africa; Solidarity Center52Tula Connell, “Mexico: App-Based Drivers Hail New Platform Law,” Solidarity Center, January 6, 2025, (<)a href='https://www.solidaritycenter.org/mexico-app-based-drivers-hail-new-platform-law-2/'(>)https://www.solidaritycenter.org/mexico-app-based-drivers-hail-new-platform-law-2(<)/a(>). and ITUC,53Nithin Coca, “Long Silenced, Gig Workers in Indonesia are Organizing and Fighting for their Rights,” (<)em(>)Equal Times(<)/em(>), March 14, 2025, (<)a href='https://www.ituc-csi.org/long-silenced-gig-workers-in'(>)https://www.ituc-csi.org/long-silenced-gig-workers-in(<)/a(>). which support global gig-worker organizing; the Data Worker’s Inquiry Project,54“Data Workers’ Inquiry,” accessed April 17, 2025, (<)a href='https://data-workers.org/'(>)https://data-workers.org/(<)/a(>). a community-based research project focusing on data labelers globally; and the Tech Equity Collaborative55“Shining a Light on Tech’s Shadow Workforce,” (<)em(>)TechEquity Contract Worker Disparity Project(<)/em(>), accessed April 17, 2025, (<)a href='https://contractwork.techequitycollaborative.org/'(>)https://contractwork.techequitycollaborative.org/(<)/a(>). and CWA’s Alphabet Workers Union,56“Dec 19, 2024 – CWA Members at Google Ratify Historic First Contract,” (<)em(>)CWA(<)/em(>), December 19, 2024, (<)a href='https://cwa-union.org/news/e-newsletter/2024-12-19'(>)https://cwa-union.org/news/e-newsletter/2024-12-19(<)/a(>). which are campaigning on temp- and contract-worker issues. 

We desperately need more ambitious and better-resourced projects. There is a particular opportunity for people who are building and training the AI systems—and, therefore, know them intimately—to use their powerful positions to hold tech firms accountable for how these systems are used. Many people working in these white-collar jobs have conveyed that their values conflict with corporate decision-making.57See, e.g., Laura Newcombe, “After Tesla Employees Signed Letter Asking Elon Musk to Resign, at Least One Has Been Fired,” (<)em(>)Gizmodo(<)/em(>), May 12, 2025, (<)a href='https://gizmodo.com/after-tesla-employees-signed-letter-asking-elon-musk-to-resign-at-least-one-has-been-fired-2000600934'(>)https://gizmodo.com/after-tesla-employees-signed-letter-asking-elon-musk-to-resign-at-least-one-has-been-fired-2000600934(<)/a(>); and Bobby Allyn, “Former Palantir Workers Condemn Company’s Work with Trump Administration,” NPR, May 5, 2025, (<)a href='https://www.npr.org/2025/05/05/nx-s1-5387514/palantir-workers-letter-trump'(>)https://www.npr.org/2025/05/05/nx-s1-5387514/palantir-workers-letter-trump.(<)/a(>) Organizing and taking collective action from these positions will have a deeply consequential impact on the trajectory of AI.

The Path Forward

During previous periods of extreme inequality, such as the Gilded Age and the Roaring Twenties, muckraking journalism, labor organizing, and strikes held unaccountable corporate power accountable, and ushered in the Progressive and New Deal eras. In our era, collective labor actions will also be critical to challenging not only Big Tech’s unchecked wealth and power over our democracy and society, but also the use of technologies to undermine job quality while reducing workers’ share of the profits they generate. Like the labor movement of the last century, today’s labor movement can fight for and win a new social compact that puts AI and digital technologies in service of the public interest and hold today’s unaccountable power accountable.

Enact a “Zero-Trust” Policy Agenda for AI

In 2023, alongside Electronic Privacy Information Center (EPIC) and Accountable Tech,58Accountable Tech et al., (<)em(>)Zero Trust AI Governance(<)/em(>), August 10, 2023, (<)a href='https://ainowinstitute.org/wp-content/uploads/2023/08/Zero-Trust-AI-Governance.pdf'(>)https://ainowinstitute.org/wp-content/uploads/2023/08/Zero-Trust-AI-Governance.pdf(<)/a(>). we recommended three pillars to guide policy action in the AI market:

  1. Time is of the essence—start by vigorously enforcing existing laws.
  2. Bold, easily administrable, bright-line rules are necessary.
  3. At each phase of the AI system life cycle, the burden should be on companies—including end users—to prove that their systems work as claimed, and are not harmful.

“Given the monumental stakes, blind trust in [tech firms’] benevolence is not an option,” we argued then. That warning rings even more starkly now.

1. Build a drumbeat of bright-line rules that restrict
the most harmful AI uses wholesale, or limit firms’
ability to collect and use data in specific ways.

Bold, bright-line rules that restrict harmful AI applications or types of data collection for AI send a clear message that the public determines whether, in what contexts, and how AI systems will be used. Compared to frameworks that rely on process-based safeguards (such as AI auditing or risk-assessment regimes) that have often, in practice, tended to further empower industry leaders and rely on robust regulatory capacity for effective enforcement,59AI Now Institute, “Algorithmic Accountability: Moving Beyond Audits,”in (<)em(>)2023 Landscape: Confronting Tech Power, AI Now Institute(<)/em(>), (<)a href='https://ainowinstitute.org/publications/algorithmic-accountability'(>)https://ainowinstitute.org/publications/algorithmic-accountability(<)/a(>) these bright-line rules have the benefit of being easily administrable and targeting the kind of harms that can’t be prevented or remedied through safeguards alone. There’s a growing list of ripe targets for these kinds of clear prohibitions. For example:60A range of bills of this variety have appeared across state legislatures. See Public Citizen, “Tracker: State Legislation on Intimate Deepfakes,” last updated April 15, 2025, (<)a href='https://www.citizen.org/article/tracker-intimate-deepfakes-state-legislation/'(>)https://www.citizen.org/article/tracker-intimate-deepfakes-state-legislation(<)/a(>); Public Citizen, “Tracker: State Legislation on Deepfakes in Elections,” last Updated April 17, 2025, (<)a href='https://www.citizen.org/article/tracker-legislation-on-deepfakes-in-elections/'(>)https://www.citizen.org/article/tracker-legislation-on-deepfakes-in-elections(<)/a(>); and California Education Code § 87359.2 (2024); H.B. 1709, 89th Leg. (Tex. 2025); H.B. 1709, 89th Leg. (Tex. 2025).

2. Regulate throughout the AI life cycle.
This means no broad-based exemptions for foundation-model providers, application developers, or end users.

AI must be regulated throughout the entire life cycle of development, from how data is collected through the training process, to fine tuning and application development, to deployment. Appropriate actors should be held responsible for decision-making at each of these stages, and be mandated to provide transparency and disclosures that enable actors further downstream to uphold their own obligations.61Cybersecurity & Infrastructure Security Agency, “CISA and UK NCSC Unveil Joint Guidelines for Secure AI System Development,” November 26, 2023, (<)a href='https://www.cisa.gov/news-events/alerts/2023/11/26/cisa-and-uk-ncsc-unveil-joint-guidelines-secure-ai-system-development'(>)https://www.cisa.gov/news-events/alerts/2023/11/26/cisa-and-uk-ncsc-unveil-joint-guidelines-secure-ai-system-development(<)/a(>).

Figure 1: Risk origination and risk proliferation in the foundation models supply chain

Fig 1: Risk Origination and Risk Proliferation in the Foundation Models Supply Chain. Source: Ada Lovelace Institute, 2023

Transparency regimes are the foundation of effective regulation of technology—necessary, but never enough62Chloe Xiant, “OpenAI’s GPT-4 Is Closed Source and Shrouded in Secrecy,” (<)em(>)Vice(<)/em(>), March 16, 2023,(<)a href='https://www.vice.com/en/article/ak3w5a/openais-gpt-4-is-closed-source-and-shrouded-in-secrecy'(>) https://www.vice.com/en/article/ak3w5a/openais-gpt-4-is-closed-source-and-shrouded-in-secrecy(<)/a(>).—and yet in the past few years we’ve seen fierce and largely successful resistance to baseline disclosure requirements from the AI industry. That’s true at the general-purpose stage where industry has used copyright and commercial secrecy as a shield to avoid disclosing training data;63Kyle Wiggers, “Many companies won’t say if they’ll comply with California’s AI training transparency law,” (<)em(>)TechCrunch(<)/em(>), October 4, 2024, (<)a href='https://techcrunch.com/2024/10/04/many-companies-wont-say-if-theyll-comply-with-californias-ai-training-transparency-law'(>)https://techcrunch.com/2024/10/04/many-companies-wont-say-if-theyll-comply-with-californias-ai-training-transparency-law(<)/a(>). but it’s equally true at the final stage, where these systems are deployed. Proactive transparency and disclosure requirements (where the consumer doesn’t need to individually request this information)64Matt Scherer, “FAQ on Colorado’s Consumer Artificial Intelligence Act (SB 24-205),” Center for Democracy and Technology, December 17, 2024, (<)a href='https://cdt.org/insights/faq-on-colorados-consumer-artificial-intelligence-act-sb-24-205/#h-the-law-requires-companies-to-give-some-information-to-consumers-proactively-without-the-consumer-having-to-request-it-why-is-that-proactive-disclosure-necessary'(>)https://cdt.org/insights/faq-on-colorados-consumer-artificial-intelligence-act-sb-24-205/#h-the-law-requires-companies-to-give-some-information-to-consumers-proactively-without-the-consumer-having-to-request-it-why-is-that-proactive-disclosure-necessary(<)/a(>). have prompted intense backlash from industry, most recently by asserting that disclosure provisions in Colorado’s AI law, passed in 2024, placed an “undue burden” on businesses.65Tamara Chuang, “Attempt to Tweak Colorado’s Controversial, First-in-the-Nation Artificial Intelligence Law is Killed,” (<)em(>)Colorado Sun(<)/em(>), May 5, 2025, (<)a href='https://coloradosun.com/2025/05/05/colorado-artificial-intelligence-law-killed/'(>)https://coloradosun.com/2025/05/05/colorado-artificial-intelligence-law-killed(<)/a(>). What this means is that in many contexts where AI is being used, people whom an AI system affects will never be made aware of its role in making determinations that can have significant consequences for their lives—let alone give them the ability to contest the system.

Industry has also exerted a concerted lobbying effort to put all of the onus on end developers, while exempting the developers of so-called “general purpose” systems.66Matt Perault, “Regulate AI Use, Not AI Development,” Andreesen Horowitz, January 27, 2025, (<)a href='https://a16z.com/regulate-ai-use-not-ai-development/'(>)https://a16z.com/regulate-ai-use-not-ai-development(<)/a(>); Jacob Helberg, “11 Elements of American AI Dominance,” (<)em(>)Republic(<)/em(>), July 26, 2024, (<)a href='https://therepublicjournal.com/journal/11-elements-of-american-ai-supremacy/'(>)https://therepublicjournal.com/journal/11-elements-of-american-ai-supremacy(<)/a(>). In 2021, alongside more than fifty global AI experts, AI Now published a letter urging the European Commission to reject this approach in the context of the EU AI Act.67AI Now Institute, “General Purpose AI Poses Serious Risks, Should Not Be Excluded From the EU’s AI Act,” policy brief, April 13, 2023, (<)a href='https://ainowinstitute.org/publications/gpai-is-high-risk-should-not-be-excluded-from-eu-ai-act'(>)https://ainowinstitute.org/publications/gpai-is-high-risk-should-not-be-excluded-from-eu-ai-act(<)/a(>). It’s important that AI be evaluated at every stage of the life cycle. Certain aspects of AI models will be best evaluated at the application stage: This is where pre-deployment testing and validation will give the most insight into the efficacy and safety of particular uses of AI.68Anna Lenhart and Sarah Myers West, (<)em(>)Lessons from the FDA for AI, AI Now Institute(<)/em(>), August 1, 2024, (<)a href='https://ainowinstitute.org/publications/research/lessons-from-the-fda-for-ai'(>)https://ainowinstitute.org/publications/research/lessons-from-the-fda-for-ai(<)/a(>). But the developers of foundation models and those involved in constructing datasets should not be exempted from scrutiny—particularly since they are generally the largest and best-resourced actors in the market,69Amba Kak and Sarah Myers West, “General Purpose AI Poses Serious Risks, Should Not Be Excluded from the EU’s AI Act,” policy report, AI Now Institute, April 13, 2023, (<)a href='https://ainowinstitute.org/publication/gpai-is-high-risk-should-not-be-excluded-from-eu-ai-act'(>)https://ainowinstitute.org/publication/gpai-is-high-risk-should-not-be-excluded-from-eu-ai-act(<)/a(>). and have demonstrated a growing propensity for opacity, offering diminishingly few details about how they train their systems.70See Margaret Mitchell et al., “Model Cards for Model Reporting,” FACCT, January 29, 2019, (<)a href='https://doi.org/10.1145/3287560.3287596'(>)https://doi.org/10.1145/3287560.3287596(<)/a(>); Jessica Ji, “What Does AI Red-Teaming Actually Mean?” Center for Security and Emerging Technology (CSET), October 24, 2023, ​​(<)a href='https://cset.georgetown.edu/article/what-does-ai-red-teaming-actually-mean/'(>)https://cset.georgetown.edu/article/what-does-ai-red-teaming-actually-mean(<)/a(>); and Heidy Khlaaf and Sarah Myers West, “Safety Co-Option and Compromised National Security: The Self-Fulfilling Prophecy of Weakened AI Risk Thresholds,” April 21, 2025, (<)a href='https://arxiv.org/abs/2504.15088'(>)https://arxiv.org/abs/2504.15088(<)/a(>). Introducing stronger accountability requirements at the foundation-model layer would motivate restructuring of the development practices, and potentially the operating model, of AI developers. In and of itself, these modes would create greater internal transparency and accountability within AI firms that would convey societal benefits, and aid the work of enforcement agencies when they need to investigate AI companies. Regulatory structures that place the onus solely on the application stage ignore that these developers, as well as purchasers and deployers, may lack the information necessary to assure compliance with existing law, absent upstream intervention.71Lenhart and West, (<)em(>)Lessons from the FDA for AI(<)/em(>). And the end users of AI systems are in many instances not consumers; they’re businesses or public agencies, themselves in need of scrutiny for the way their use of the system affects consumers, workers, and members of the public.

3. The AI industry shouldn’t be grading
its own homework: Oversight of AI model
development must be independent.

Industry tends to rely too heavily on “red-teaming,” “model cards,” and similar approaches to system evaluation, where companies are the primary actors responsible for determining whether their systems work as intended and for mitigating harm.72OpenAI, “Advancing Red Teaming With People and AI,” November 21, 2024, (<)a href='https://openai.com/index/advancing-red-teaming-with-people-and-ai/'(>)https://openai.com/index/advancing-red-teaming-with-people-and-ai(<)/a(>); Microsoft AI Red Team, (<)em(>)Lessons from Red Teaming 100 Generative AI Products(<)/em(>), Microsoft, 2024, (<)a href='https://ashy-coast-00aeb501e.6.azurestaticapps.net/MS_AIRT_Lessons_eBook.pdf'(>)https://ashy-coast-00aeb501e.6.azurestaticapps.net/MS_AIRT_Lessons_eBook.pdf(<)/a(>); Rebecca Pifer, “CHAI Launches Registry for Health AI Model Cards,” (<)em(>)Healthcare Dive(<)/em(>), February 28, 2025, (<)a href='https://www.healthcaredive.com/news/coalition-health-ai-model-card-registry/741037/'(>)https://www.healthcaredive.com/news/coalition-health-ai-model-card-registry/741037(<)/a(>). This approach has many problems; it ignores, for example, the significant conflicts of interest involved and fails to employ robust methodologies.73Khlaaf and West, “Safety Co-Option and Compromised National Security.” 

Companies have asserted that they hold the unique technical expertise necessary to evaluate their own systems74“Fmr. Google CEO Says No One in Government Can Get AI Regulation ‘Right’,” NBC News, May 15, 2023, (<)a href='https://www.nbcnews.com/meet-the-press/video/fmr-google-ceo-says-no-one-in-government-can-get-ai-regulation-right-174442053869'(>)https://www.nbcnews.com/meet-the-press/video/fmr-google-ceo-says-no-one-in-government-can-get-ai-regulation-right-174442053869(<)/a(>). in order to execute what is essentially a fully self-regulatory approach to governance, making a set of “voluntary commitments” to the Biden administration in 2023.75White House, “FACT SHEET: Biden-Harris Administration Secures Voluntary Commitments from Eight Additional Artificial Intelligence Companies to Manage the Risks Posed by AI,” September 12, 2023, (<)a href='https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/09/12/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-eight-additional-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai/'(>)https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/09/12/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-eight-additional-artificial-intelligence-companies-to-manage-the-risks-posed-by-ai(<)/a(>). When asked a year later about their progress on those commitments, the response was milquetoast: While companies reported taking actions like engaging in internal testing to mitigate cybersecurity risks, they were unable to validate whether any of those actions actually reduced underlying risk; and by focusing on domains like bioweapons, they ignored the wider universe of harms like consumer and worker exploitation, privacy, and copyright.76Melissa Heikkilä, “AI Companies Promised to Self-Regulate One Year Ago. What’s Changed?” (<)em(>)Technology Review(<)/em(>), July 22, 2024, (<)a href='https://www.technologyreview.com/2024/07/22/1095193/ai-companies-promised-the-white-house-to-self-regulate-one-year-ago-whats-changed/'(>)https://www.technologyreview.com/2024/07/22/1095193/ai-companies-promised-the-white-house-to-self-regulate-one-year-ago-whats-changed(<)/a(>). And OpenAI stepped back from even fulfilling the basic obligations for these commitments in the release of GPT-4.1, opting not to release a model card on the grounds that the model “isn’t at the frontier.”77Maxwell Zeff, “OpenAI Ships GPT-4.1 Without a Safety Report,” (<)em(>)TechCrunch(<)/em(>), April 15, 2025, (<)a href='https://techcrunch.com/2025/04/15/openai-ships-gpt-4-1-without-a-safety-report/'(>)https://techcrunch.com/2025/04/15/openai-ships-gpt-4-1-without-a-safety-report(<)/a(>). 

This illustrates a point that by now should be abundantly clear: Companies cannot be responsible for their own self-evaluation, and are poorly incentivized to uphold any promises they make absent the foundation of enforceable law:

4. Create remedies that strike
at the root of AI power: data.

We don’t need to limit ourselves to the standard AI policy tool kit: A wide range of creative remedies could be deployed by enforcement agencies to strike at the root of AI firms’ power. These include measures such as:

5. We need a sharp-edged competition
tool kit to de-rig the AI market, and how
AI shapes market power across sectors.

Amid concerns of Big Tech dominance in AI growing more prominent in a frenzied market, competition enforcers globally have been far from silent. Over the past few years, we have seen multiple early inquiries into so-called partnership agreements between AI startups and Big Tech hyperscalers;84Federal Trade Commission, (<)em(>)Partnerships Between Cloud Service Providers and AI Developers(<)/em(>), staff report, January 2025, (<)a href='https://www.ftc.gov/system/files/ftc_gov/pdf/p246201_aipartnerships6breport_redacted_0.pdf'(>)https://www.ftc.gov/system/files/ftc_gov/pdf/p246201_aipartnerships6breport_redacted_0.pdf(<)/a(>); European Commission, “Commission Launches Call for Contributions on Competition in Virtual Worlds and Generative AI,” press release, January 8, 2024, (<)a href='https://ec.europa.eu/commission/presscorner/detail/en/ip_24_85'(>)https://ec.europa.eu/commission/presscorner/detail/en/ip_24_85(<)/a(>); Competition and Markets Authority, “Microsoft / OpenAI Partnership Merger Inquiry,” Gov.UK, April 15, 2025, (<)a href='https://www.gov.uk/cma-cases/microsoft-slash-openai-partnership-merger-inquiry'(>)https://www.gov.uk/cma-cases/microsoft-slash-openai-partnership-merger-inquiry(<)/a(>). several market studies into concentration in the cloud market, from Japan to the Netherlands to the UK to the US;85Japan Fair Trade Commission, “Report Regarding Cloud Services,” press release, June 28, 2022, (<)a href='https://www.jftc.go.jp/en/pressreleases/yearly-2022/June/220628.html'(>)https://www.jftc.go.jp/en/pressreleases/yearly-2022/June/220628.html(<)/a(>); Authority for Consumers and Markets, “Market Study into Cloud Services,” May 9, 2022, (<)a href='https://www.acm.nl/en/publications/market-study-cloud-services'(>)https://www.acm.nl/en/publications/market-study-cloud-services(<)/a(>); Competition and Markets Authority, (<)em(>)Cloud Services Market Investigation(<)/em(>), January 28, 2025, (<)a href='https://assets.publishing.service.gov.uk/media/67989251419bdbc8514fdee4/summary_of_provisional_decision.pdf'(>)https://assets.publishing.service.gov.uk/media/67989251419bdbc8514fdee4/summary_of_provisional_decision.pdf(<)/a(>); Nick Jones, “Cloud Computing FRI: What We Heard and Learned,” FTC(<)em(>) (<)/em(>)(blog), November 16, 2023, (<)a href='https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2023/11/cloud-computing-rfi-what-we-heard-learned'(>)https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2023/11/cloud-computing-rfi-what-we-heard-learned(<)/a(>). and detailed reports highlighting potential choke points and competition risks across the AI supply chain.86Competition and Markets Authority, (<)em(>)AI Foundation Models Update Paper(<)/em(>), April 11, 2024, (<)a href='https://assets.publishing.service.gov.uk/media/661941a6c1d297c6ad1dfeed/Update_Paper__1_.pdf'(>)https://assets.publishing.service.gov.uk/media/661941a6c1d297c6ad1dfeed/Update_Paper__1_.pdf(<)/a(>); Office of Technology, “Tick, Tick, Tick. Office of Technology’s Summit on AI,” FTC, January 18, 2024, (<)a href='https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/01/tick-tick-tick-office-technologys-summit-ai'(>)https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/01/tick-tick-tick-office-technologys-summit-ai(<)/a(>). The DOJ, FTC, CMA, and EC even issued a joint statement on competitive concerns with AI, warning that partnerships and investments could be used to undermine competition.87Federal Trade Commission, “FTC, DOJ, and International Enforcers Issue Joint Statement on AI Competition Issues,” press release, July 23, 2024, (<)a href='https://www.ftc.gov/news-events/news/press-releases/2024/07/ftc-doj-international-enforcers-issue-joint-statement-ai-competition-issues'(>)https://www.ftc.gov/news-events/news/press-releases/2024/07/ftc-doj-international-enforcers-issue-joint-statement-ai-competition-issues(<)/a(>). This is already a promising departure from the past decade of a passive stance toward the consumer tech industry, which has left regulators struggling to catch up to remedy harm—from Facebook’s acquisitions of WhatsApp and Instagram88Laya Neelakandan, “Meta Resorted to ‘Buy-or-Bury Scheme’ With Instagram and WhatsApp Deals, Former FTC Chair Lina Khan Says,” CNBC, April 14, 2025, (<)a href='https://www.cnbc.com/2025/04/14/former-ftc-chair-khan-meta-acquisitions-instagram-whatsapp.html'(>)https://www.cnbc.com/2025/04/14/former-ftc-chair-khan-meta-acquisitions-instagram-whatsapp.html(<)/a(>). to Google’s sprawling search monopoly.89Michael Liedtke, “Google’s Digital Ad Network Declared an Illegal Monopoly, Joining Its Search Engine in Penalty Box,” Associated Press, April 17, 2025, (<)a href='https://apnews.com/article/google-illegal-monopoly-advertising-search-a1e4446c4870903ed05c03a2a03b581e'(>)https://apnews.com/article/google-illegal-monopoly-advertising-search-a1e4446c4870903ed05c03a2a03b581e(<)/a(>). 

Yet, despite many of the “right noises,” much of this activity failed to materialize into concrete enforcement action and legislative change, or to draw bright lines prohibiting specific anticompetitive business practices. A familiar dynamic has consisted of limited and restrictive interpretations of existing legal standards for enforcement action. For example, traditional merger review conditions are unable to capture more indirect means of dependence and control between AI startups and cloud hyperscalers. Or consider the threshold in the US’s Hart-Scott-Rodino Act, which meant that smaller acquisitions by Big Tech firms were not subject to scrutiny, like Rockset (acquired by OpenAI), Deci AI (acquired by NVIDIA), OctoAI (also acquired by NVIDIA). Across these issues persists a more fundamental limitation: the artificial blinkers that prevent regulators from fully grappling with the many dimensions of power and control that Big Tech firms exercise through their sprawling ecosystems. A recent report from the Competition and Markets Authority (CMA) in the UK published a diagram (see Fig 2) depicting the presence of Amazon, Apple, Google, Meta, and Microsoft across the AI supply chain—illuminating the ways in which their presence across these markets allows for network effects, data feedback loops, and unparalleled access to monetizable markets for their AI products.90Competition and Markets Authority, (<)em(>)AI Foundation Models Technical Update Report(<)/em(>), April 16, 2024, (<)a href='https://assets.publishing.service.gov.uk/media/661e5a4c7469198185bd3d62/AI_Foundation_Models_technical_update_report.pdf'(>)https://assets.publishing.service.gov.uk/media/661e5a4c7469198185bd3d62/AI_Foundation_Models_technical_update_report.pdf(<)/a(>). (We canvass these in Chapter 2: Heads I Win, Tails You Lose.)

Fig 2: How Amazon, Apple, Google, Meta, and Microsoft permeate the AI supply chain. Source: CMA; Graphic Design: Mozilla Foundation.

Yet the irony is that the CMA, among other counterparts, has fallen short of acting in ways that might interrupt this ecosystem control. The scope for intervention is limited as much by restrictive legal thresholds as it is by a corporate lobbying environment around AI that pits any market intervention not only as anti-innovation (see Chapter 1.4: Recasting Regulation), but as fundamentally harmful to the national interest given the stakes of the so-called AI arms race (see Chapter 1.3: AI Arms Race 2.0). Staying with the UK example, Microsoft announced a massive investment in compute on the heels of two significant enforcement actions affecting Microsoft: the CMA’s decision not to pursue a block of the Microsoft-Activision merger, and the opening of an investigation by the CMA into competition concerns in the cloud market. AI firms use investment as a clear play to hedge against the very real risk that the CMA will take action in ways that would threaten the companies’ bottom line.

Looking ahead, we see an opportunity to corral around a set of targeted interventions that could tackle Big Tech’s control over the broader tech ecosystem, and could go a long way in shifting the “rigged” dynamics of the AI market (see Chapter 2: Heads I Win, Tails You Lose). We spotlight some key policy directions here that identify high-value intervention points.91States might be ideal testing grounds here, especially as many of these moves might require legislative changes. A recent report by the Institute for Local Self Reliance identifies state antitrust laws as being ripe for enforcement and legislative reform—pointing to enforcement victories in Washington, Colorado, Arizona, and New York, among other states——as well as successful attempts at passing targeted legislative reforms giving these enforcers better tools to grapple with anticompetitive conduct. See Ron Knox, (<)em(>)The State(s) of Antitrust: How States Can Strengthen Monopoly Laws and Get Enforcement Back on Track(<)/em(>), Institute for Local Self-Reliance, April 2025, (<)a href='https://ilsr.org/wp-content/uploads/2025/03/ILSR-StatesofAntitrust.pdf'(>)https://ilsr.org/wp-content/uploads/2025/03/ILSR-StatesofAntitrust.pdf(<)/a(>); and Knox, “Mapping Our State Antitrust Laws,” Institute for Local Self-Reliance, April 1, 2025, (<)a href='https://ilsr.org/articles/state-antitrust-law-database-and-map/'(>)https://ilsr.org/articles/state-antitrust-law-database-and-map(<)/a(>).

6. Rigorous oversight for new data
center development and operation

As AI spurs the growth of data center development across the world and projects staggering demands for power, policymakers must reckon with the incompatibility between this unrestrained growth and our already resource-constrained environments. Communities near data centers face the realities of air pollution, lack of adequate water, and noise pollution92See Alli Finn, (<)em(>)The Costs of Data Centers to Our Communities—and How to Fight Back(<)/em(>), Kairos, 2025, 8–10, (<)a href='https://www.kairosfellows.org/fightdatacenters'(>)https://www.kairosfellows.org/fightdatacenters(<)/a(>); More Perfect Union, “I Live 400 Yards From Mark Zuckerberg’s Massive Data Center,” YouTube video, March 27, 2025, (<)a href='https://www.youtube.com/watch?v=DGjj7wDYaiI&ab_channel=MorePerfectUnion'(>)https://www.youtube.com/watch?v=DGjj7wDYaiI&ab_channel=MorePerfectUnion(<)/a(>). with a disproportionate impact on Black and Brown frontline communities.93Adam Mahoney, “America’s Digital Demand Threatens Black Communities with More Pollution,” (<)em(>)Capital B(<)/em(>), February 25, 2025, (<)a href='https://capitalbnews.org/ai-data-centers-south-carolina-black-communities/'(>)https://capitalbnews.org/ai-data-centers-south-carolina-black-communities(<)/a(>); and Dwaign Tyndal, “The Dark Side of AI: Climate Chaos, Pollution, and Injustice,” (<)em(>)CommonWealth Beacon(<)/em(>), February 3, 2025, (<)a href='https://commonwealthbeacon.org/opinion/the-dark-side-of-ai-climate-chaos-pollution-and-injustice/'(>)https://commonwealthbeacon.org/opinion/the-dark-side-of-ai-climate-chaos-pollution-and-injustice(<)/a(>). Yet data centers don’t merely affect those who live closest to them: To fund the massive investment required to build out the infrastructure necessary to fuel AI data centers (such as power plants and transmission lines), energy utility companies may pass off the costs onto ordinary ratepayers.94Eliza Martin and Ari Peskoe, (<)em(>)Extracting Profits from the Public: How Utility Ratepayers are Paying for Big Tech’s Power(<)/em(>), Harvard Electricity Law Initiative, March 5, 2025, (<)a href='https://eelp.law.harvard.edu/wp-content/uploads/2025/03/Harvard-ELI-Extracting-Profits-from-the-Public.pdf'(>)https://eelp.law.harvard.edu/wp-content/uploads/2025/03/Harvard-ELI-Extracting-Profits-from-the-Public.pdf(<)/a(>). Already Americans are seeing their electricity rates increase;95Dave Williams, “Georgia Senate Committee Passes Bill to Protect Residents From Data Center Costs,” (<)em(>)Current(<)/em(>), February 25, 2025, (<)a href='https://thecurrentga.org/2025/02/25/georgia-senate-committee-passes-bill-to-protect-residents-from-data-center-costs/'(>)https://thecurrentga.org/2025/02/25/georgia-senate-committee-passes-bill-to-protect-residents-from-data-center-costs(<)/a(>); Josh Janey, “Dominion Energy Proposes Hefty Rate Hikes,” (<)em(>)Virginia Business(<)/em(>), April 1, 2025, (<)a href='https://virginiabusiness.com/dominion-energy-proposes-hefty-rate-hikes/'(>)https://virginiabusiness.com/dominion-energy-proposes-hefty-rate-hikes(<)/a(>); and Citizens Action Coalition, “Duke Energy Rate Hike (2024), accessed May 5, 2025, (<)a href='https://www.citact.org/duke-rate-hike-2024'(>)https://www.citact.org/duke-rate-hike-2024(<)/a(>). this is only expected to intensify. 

Meanwhile, states and localities are handing out billions of dollars in corporate tax subsidies to data center developers by exempting them from paying sales, use, and property taxes, or by discounting their rates.96See Kasia Tarczynska and Greg LeRoy, (<)em(>)Cloudy With a Loss of Spending Control: How Data Centers Are Endangering State Budgets(<)/em(>), Good Jobs First, April 2025, (<)a href='https://goodjobsfirst.org/cloudy-with-a-loss-of-spending-control-how-data-centers-are-endangering-state-budgets/'(>)https://goodjobsfirst.org/cloudy-with-a-loss-of-spending-control-how-data-centers-are-endangering-state-budgets(<)/a(>); Good Jobs First, “Subsidy Tracker,” accessed May 18, 2025, (<)a href='https://subsidytracker.goodjobsfirst.org/'(>)https://subsidytracker.goodjobsfirst.org(<)/a(>); and Finn, (<)em(>)The Costs of Data Centers to Our Communities(<)/em(>). This comes directly at the expense of communities, who lose out on tax revenue for local schools, infrastructure, and other community needs. 

Despite billions of dollars of investment and government subsidies, much of the current data center development process operates in obscurity.97Eric Bonds and Viktor Newby, “Data Centers, Non-Disclosure Agreements and Democracy,” (<)em(>)Virginia Mercury(<)/em(>), April 30, 2025, (<)a href='https://virginiamercury.com/2025/04/30/data-centers-non-disclosure-agreements-and-democracy/'(>)https://virginiamercury.com/2025/04/30/data-centers-non-disclosure-agreements-and-democracy(<)/a(>). Many data centers fail to report even basic information about their resource consumption, including power usage and water consumption. This prevents states from planning for adequate supply and demand, threatening the stability of the grid. Moreover, many deals between data centers and utility companies are approved by regulators in proceedings that tend to go uncontested by the public, preventing ratepayers from understanding the impacts these large customers may have on their rates.98See Issie Lapowsky, “As Data Centers Push Into States, Lawmakers Are Pushing Right Back,” (<)em(>)Tech Policy Press(<)/em(>), March 24, 2025, (<)a href='https://www.techpolicy.press/as-data-centers-push-into-states-lawmakers-are-pushing-right-back/'(>)https://www.techpolicy.press/as-data-centers-push-into-states-lawmakers-are-pushing-right-back(<)/a(>); and Martin and Peskoe, (<)em(>)Extracting Profits from the Public(<)/em(>).

Yet this current path of unrestrained data center growth is not inevitable. More communities are realizing that successful economic development cannot hinge on environmental, health, and economic harms: Recently, two counties in Atlanta have temporarily paused all new data center projects to adequately ensure that development protects community interests;99Zachary Hansen, “$17B Data Center Proposal in Metro Atlanta is Latest to Prompt Moratorium,” (<)em(>)Atlanta Journal-Constitution(<)/em(>), May 8, 2025, (<)a href='https://www.ajc.com/news/business/coweta-county-issues-moratorium-to-pause-data-center-projects/2J464ISZIVB4VMFSBXZXNVHZXE/'(>)https://www.ajc.com/news/business/coweta-county-issues-moratorium-to-pause-data-center-projects/2J464ISZIVB4VMFSBXZXNVHZXE(<)/a(>). and a county board of supervisors in Virginia rejected a plan for a gas power plant and data center campus because of concerns about the increased rates of pollution and negative environmental impact it would cause in the surrounding area.100Matthew Gooding, “Plan for Gas-Powered Data Center Campus in Pittsylvania County, Virginia, Withdrawn Again,” Data Center Dynamics, April 15, 2025, (<)a href='https://www.datacenterdynamics.com/en/news/plan-for-gas-powered-data-center-campus-in-pittsylvania-county-virginia-withdrawn-again/'(>)https://www.datacenterdynamics.com/en/news/plan-for-gas-powered-data-center-campus-in-pittsylvania-county-virginia-withdrawn-again(<)/a(>). The developer is now considering building housing instead.

Policymakers must take swift and decisive action to ensure that data center development supports, rather than harms, the needs of local communities and the resources available on our planet, such as the following:

Bridge Networks of Expertise, Policy Domains, and Narrative 

Blinkered views that fail to see the different components of the AI supply chain materially and are often single-issue focused have too often undermined AI advocacy and policy. It is easy to miss the ways in which big-picture narratives (see Chapters 1.1-1.4 on AI’s False Gods) manifest on the ground across policy fights. Silos can blunt strategic creativity in ways that foreclose alliance or cross-pollination. In our 2023 report, we attempted to find ways to identify and bridge some of these siloes: like forming a united strategy around privacy and competition policy; or understanding industrial policy as central to (not distinct from) discourse on AI regulation.112Amba Kak and Sarah Myers West, (<)em(>)2023 Landscape: Confronting Tech Power(<)/em(>), AI Now Institute, April 11, 2023, (<)a href='https://ainowinstitute.org/publications/research/2023-landscape-confronting-tech-power'(>)https://ainowinstitute.org/publications/research/2023-landscape-confronting-tech-power(<)/a(>). Here we highlight additional emergent areas of cross-pollination in a bid to bolster and strengthen our strategies:

1. National Security as a Vector Both
For and Against AI Accountability

National security arguments are a key vector driving the AI infrastructure acceleration agenda. Equally, national security arguments support moves for stronger privacy, security, and accuracy thresholds, as well as moves that combat concentrated infrastructure markets as threatening resilience.

As we detail in Chapter 1.3 on AI & Geopolitics, recent national security policy proposals (under President Biden and now President Trump) and the specter of Chinese AI competition have consistently cleared the pathway for the industry to expand through infrastructure build-out.113See also AI Now Institute, “Tracking the US and China AI Arms Race,” April 11, 2023, (<)a href='https://ainowinstitute.org/publications/tracking-the-us-and-china-ai-arms-race'(>)https://ainowinstitute.org/publications/tracking-the-us-and-china-ai-arms-race(<)/a(>). Most recently, baseline transparency mechanisms around data center build-out in Virginia’s HB 2035 bill,114H.B. 2035, Reg. Sess. (Va. 2025). championed by local advocacy groups, were shut down citing national security concerns, including alleged pushback from Homeland Security.115Because H.B. 2035 was stricken from the docket, it was never discussed on the record. However, the Senate version of the bill is instructive: The original Senate bill closely mirrors H.B. 2035. S.B. 1353, Reg. Sess. (Va. 2025) (Introduced), (<)a href='https://lis.blob.core.windows.net/files/1019728.PDF'(>)https://lis.blob.core.windows.net/files/1019728.PDF(<)/a(>). The bill’s substitute version, however, directly discusses national security concerns. See S.B. 1353, Reg. Sess. (Va. 2025) (Senate Substitute), (<)a href='https://lis.blob.core.windows.net/files/1034365.PDF'(>)https://lis.blob.core.windows.net/files/1034365.PDF(<)/a(>).

On the other hand, national security imperatives (resilience, reliability, security) can be a strategic vector to demand more accountability from the AI industry, and an opportunity to bring security experts into broader debates about AI accountability. Former military officials and security experts alike have sounded the alarm on inherent vulnerabilities in using commercial language models in military contexts116See examples in Heidy Khlaaf and Sarah M. West, “The Rush to A.I. Threatens National Security,” (<)em(>)New York Times(<)/em(>), January 27, 2025, (<)a href='https://www.nytimes.com/2025/01/27/opinion/ai-trump-military-national-security.html'(>)https://www.nytimes.com/2025/01/27/opinion/ai-trump-military-national-security.html(<)/a(>).—and while these intractable concerns (flaws in large-scale AI are features, not bugs, as we note in Chapter 3) are unconscionable in high-stakes combat environments, they are equally worrisome for many other swaths of applications in the economy and society.

Anti-monopoly groups are also increasingly foregrounding structural vulnerabilities created by AI infrastructure that is concentrated among very few companies, and across different points of the AI supply chain.117Former FTC Chair Lina Khan urged skepticism toward the argument that the United States must coddle its domestic monopolies—or “national champions”—to maintain global dominance. In her remarks to the FTC Technology Summit last year, she discussed the pitfalls of policymakers shielding large corporations from competition and identified the “decline and degradation” of Boeing as an example of this historically failed strategy. “The difference between Boeing and many of [the other corporations coddled by policymakers],” she said, “is there’s simply no masking airplanes falling apart in the sky.” See Lina Khan, “Remarks of Chair Lina M. Khan FTC Tech Summit,” speech, Federal Trade Commission, January 25, 2024, (<)a href='https://www.ftc.gov/system/files/ftc_gov/pdf/2024.01.25-chair-khan-remarks-at-ot-tech-summit.pdf'(>)https://www.ftc.gov/system/files/ftc_gov/pdf/2024.01.25-chair-khan-remarks-at-ot-tech-summit.pdf(<)/a(>); and Lina Khan, “Remarks by Chair Lina M. Khan As Prepared for Delivery,” speech, Carnegie Endowment for International Peace, March 13, 2024, (<)a href='https://www.ftc.gov/system/files/ftc_gov/pdf/2024.03.13-chair-khan-remarks-at-the-carnegie-endowment-for-intl-peace.pdf'(>)https://www.ftc.gov/system/files/ftc_gov/pdf/2024.03.13-chair-khan-remarks-at-the-carnegie-endowment-for-intl-peace.pdf(<)/a(>). The creation of single points of failure—recently, an erroneous update to Microsoft’s operating system from the cybersecurity firm CrowdStrike—triggered the largest IT outage in history, affecting global businesses, hospitals, airports, courts, and governments.118Microsoft was quick to find a scapegoat by reviewing whether EU digital regulations may have come into play in the incident. See Competition Policy International, “Microsoft Investigates European Commission’s Role in Recent CrowdStrike Outage,” July 22, 2024, (<)a href='https://www.pymnts.com/cpi-posts/microsoft-investigates-european-commissions-role-in-recent-crowdstrike-outage'(>)https://www.pymnts.com/cpi-posts/microsoft-investigates-european-commissions-role-in-recent-crowdstrike-outage(<)/a(>); and Cristiano Lima-Strong and Cat Zakrzewski, “Microsoft’s Global Sprawl Comes Under Fire After Historic Outage,” (<)em(>)Washington Post(<)/em(>), July 20, 2024, (<)a href='https://www.washingtonpost.com/technology/2024/07/20/microsoft-outage-crowdstrike-regulatory-scrutiny/'(>)https://www.washingtonpost.com/technology/2024/07/20/microsoft-outage-crowdstrike-regulatory-scrutiny(<)/a(>). CrowdStrike is only the most conspicuous example of the fragility of a super-concentrated infrastructure—the AI stack is rife with choke points from chips to undersea cables—and we’re seeing the risk amplified with errors in AI foundation models that could undergird applications across sectors of the economy. Former SEC Chair Gary Gensler has noted how concentration in the market for AI models could produce conditions that might lead us to the next financial crisis.119Declan Harty and Steven Overly, “Gensler’s Warning: Unchecked AI Could Spark Future Financial Meltdown,” (<)em(>)Politico(<)/em(>), March 19, 2024, (<)a href='https://www.politico.com/news/2024/03/19/sec-gensler-artificial-intelligence-00147665'(>)https://www.politico.com/news/2024/03/19/sec-gensler-artificial-intelligence-00147665(<)/a(>). 

These vulnerabilities are more acute when it comes to the public sector, especially as a growing proportion of government runs on Big Tech cloud infrastructure.120In addition, we’re seeing AI systems used to manage and distribute public benefits and essential services. The flaws of large-scale AI models—like leakage, hallucinations, and bias (see (<)strong(>)(<)a href='/publications/research/3-consulting-the-record-ai-consistently-fails-the-public'(>)Chapter 3(<)/a(>)(<)/strong(>))—are even more dangerous in these circumstances because they risk directly undermining vital public services. See U.S. Department of Health & Human Services, “Plan for Promoting Responsible Use of Artificial Intelligence in Automated and Algorithmic Systems by State, Local, Tribal, and Territorial Governments in Public Benefit Administration,” March 28, 2024, (<)a href='https://www.hhs.gov/sites/default/files/public-benefits-and-ai.pdf'(>)https://www.hhs.gov/sites/default/files/public-benefits-and-ai.pdf(<)/a(>). This will inevitably increase manifold with the rushed push toward expanding adoption of commercially available AI tools121Heidy Khlaaf, Sarah Myers West and Meredith Whittaker. (<)em(>)Mind the Gap: Foundation Models and the Covert Proliferation of Military Intelligence, Surveillance, and Targeting. (<)/em(>)AI Now Institute, October 18, 2024, (<)a href='https://arxiv.org/abs/2410.14831'(>)https://arxiv.org/abs/2410.14831(<)/a(>). (via DOGE’s efforts, among other preexisting initiatives such as the Department of Defense’s Task Force Lima).122U.S. Department of Defense, “DOD Announces Establishment of Generative AI Task Force,” press release, August 10, 2023, (<)a href='https://www.defense.gov/News/Releases/Release/Article/3489803/dod-announces-establishment-of-generative-ai-task-force/'(>)https://www.defense.gov/News/Releases/Release/Article/3489803/dod-announces-establishment-of-generative-ai-task-force(<)/a(>). Hyperscalers have done their part to market directly to government services, touting their security credentials,123See Amazon, “The Trusted Cloud for Government,” accessed May 18, 2025, (<)a href='https://aws.amazon.com/government-education/government/'(>)https://aws.amazon.com/government-education/government(<)/a(>); and Karen Dahut and Phil Venables, “How to Build a Secure Foundation for American Leadership in AI,” Google, October 18, 2023, (<)a href='https://blog.google/technology/ai/how-to-build-a-secure-foundation-for-american-leadership-in-ai/'(>)https://blog.google/technology/ai/how-to-build-a-secure-foundation-for-american-leadership-in-ai(<)/a(>). but have underplayed the long-term implications of introducing single points of failure into national security infrastructure. In July 2023, Microsoft revealed a breach by Chinese state-backed hackers who accessed emails of top US officials, following a previous security lapse with Russian-backed SolarWinds attackers accessing government data.124Associated Press, “Scathing Federal Report Rips Microsoft for Response to Chinese Hack,” Voice of America (VOA), April 3, 2024, (<)a href='https://www.voanews.com/a/scathing-federal-report-rips-microsoft-for-response-to-chinese-hack/7556113.html'(>)https://www.voanews.com/a/scathing-federal-report-rips-microsoft-for-response-to-chinese-hack/7556113.html(<)/a(>). Addressing these serious security concerns would have beneficial implications for the use of AI across the wider economy. 

2. Privacy as an Economic Justice Issue:
Disrupting the Surveillance-to-Automation Pipeline

Whether it’s Hollywood actors resisting invasive body scans as a means to create digital replicas of themselves that might eventually replace them,125Pranshu Verma, “Digital Clones Made by AI Tech Could Make Hollywood Extras Obsolete,” (<)em(>)Washington Post(<)/em(>), July 19, 2023, (<)a href='https://www.washingtonpost.com/technology/2023/07/19/ai-actors-fear-sag-strike-hollywood/'(>)https://www.washingtonpost.com/technology/2023/07/19/ai-actors-fear-sag-strike-hollywood(<)/a(>). or the uptick in the market for wearable surveillance tools in the workplace that will then be used to train “humanoid robots,”126Jacob Biba, “Top 27 Humanoid Robots in Use Right Now,” Built In, updated March 10, 2025, (<)a href='https://builtin.com/robotics/humanoid-robots'(>)https://builtin.com/robotics/humanoid-robots(<)/a(>). the connections between surveillance and the eventual displacement or devaluation of workers is becoming increasingly explicit. The US and UK governments, meanwhile, are making claims that they can train AI systems on government workflows that will automate large parts of government work.127See for example Jack Aldane and Richard Johnstone, “UK Deputy PM Says Embracing AI ‘Will Reduce Civil Service Headcount’, Estonia’s ‘Personal Government’ Vision, and More,” Global Government Forum, March 7, 2024, (<)a href='https://www.globalgovernmentforum.com/uk-deputy-pm-says-embracing-ai-will-reduce-civil-service-headcount-estonias-personal-government-vision-and-more/'(>)https://www.globalgovernmentforum.com/uk-deputy-pm-says-embracing-ai-will-reduce-civil-service-headcount-estonias-personal-government-vision-and-more(<)/a(>); and Matteo Wong, “DOGE’s Plans to Replace Humans with AI Are Already Under Way,” (<)em(>)Atlantic(<)/em(>), March 10, 2025, (<)a href='https://www.theatlantic.com/technology/archive/2025/03/gsa-chat-doge-ai/681987/'(>)https://www.theatlantic.com/technology/archive/2025/03/gsa-chat-doge-ai/681987(<)/a(>).

This surveillance-to-automation pipeline situates traditional privacy tools (like consent, opt-outs, impermissible purposes, and data minimization) as central levers in enabling more just economic conditions. Scholar Ifeoma Ajunwa argues that worker data should be seen as “captured capital” for firms, where workers risk having their data used to train technologies that will eventually displace them (or create the conditions to reduce their wages), or having their data sold to the highest bidder via a growing network of data brokers, without control or compensation.128Ifeoma Ajunwa, “A.I. and Captured Capital,” (<)em(>)Yale Law Journal Forum(<)/em(>) 134 (2025): (<)a href='https://www.yalelawjournal.org/forum/ai-and-captured-capital'(>)https://www.yalelawjournal.org/forum/ai-and-captured-capital(<)/a(>). Meanwhile, as the AI industry scrambles to find a conclusive business model, all trends point toward a proliferation of enterprise use cases that all appear to rely on the capture of workflow data (through the ratcheting up of workplace surveillance), which can then be used to automate more workflows.129Ben Thompson, “Enterprise Philosophy and the First Wave of AI,” Stratechery, September 24, 2024, (<)a href='https://stratechery.com/2024/enterprise-philosophy-and-the-first-wave-of-ai/'(>)https://stratechery.com/2024/enterprise-philosophy-and-the-first-wave-of-ai(<)/a(>). We’re also seeing extractive practices proliferate in the hyper-surveilled context of prisons, which are increasingly targeted by AI startups eyeing incarcerated workers as a cheap source of labor to train their AI algorithms. These “clickworkers” or “human labelers” are paid pennies on the dollar—if anything at all—by startups like Vainu, which ironically boasts that it is “creating work because of AI [to] employ and empower the new working class.” 

This necessitates reframing worker privacy as core to the economic justice agenda around AI. Specifically, it requires identifying the point of collection (i.e. , surveillance) as the appropriate target for pushback given the ways in which this will be weaponized against worker interests downstream. In regulatory terms, this could mean prioritizing data-minimization rules that restrain the collection of certain kinds of worker data, or restrict the use of worker data collected in one context for training AI. Thinking around data privacy will also need to expand to meet the moment —revising narrow notions of what counts as personal information, and prioritizing structural and absolute constraints on workplace surveillance that don’t rely on consent-based mechanisms. 

State advocacy groups are already prioritizing these issues,130Tyler Katzenberger and Chase DiFeliciantonio, “Labor Channels George Orwell on AI,” (<)em(>)Politico(<)/em(>), March 4, 2025, (<)a href='https://www.politico.com/newsletters/politico-technology-california-decoded-preview/2025/03/04/labor-channels-george-orwell-on-ai-211284'(>)https://www.politico.com/newsletters/politico-technology-california-decoded-preview/2025/03/04/labor-channels-george-orwell-on-ai-211284(<)/a(>). including in the following proposals:

Likewise, existing law applies to AI use cases:

3. Privacy and Competition in the Age of Generative AI:
Data Minimization to Interrupt Data Moats

In our 2023 report, we warned that privacy, competition expertise, and tools were often siloed from one another, leading to interventions that easily compromise the objectives of one issue over the other.139Kak and West, (<)em(>)2023 Landscape: Confronting Tech Power(<)/em(>). Firms continue to take advantage of this to amass information asymmetries that contribute to a further concentration of their power. We spotlighted the need for heightened scrutiny on “data mergers”140AI Now Institute, “Toxic Competition: Regulating Big Tech’s Data Advantage,” in (<)em(>)2023 Landscape: Confronting Tech Power(<)/em(>), April 11, 2023, (<)a href='https://ainowinstitute.org/publications/toxic-competition'(>)https://ainowinstitute.org/publications/toxic-competition(<)/a(>).—accounting for how firms leverage commercial surveillance tools and strategies to amass power—as well as adtech rules restricting first-party data collection for advertising purposes as having twin privacy and competition benefits. 

This cross-pollination is even more important now: Data is a key bottleneck in the development of generative AI, and a strategic point of intervention for both protecting privacy and limiting Big Tech from entrenching its data advantages in the AI market. The idea that freely available, publicly accessible data is “everywhere” and therefore not a scarce resource is intuitively appealing but misses the point. Datasets with high levels of curation and human feedback, and those that come with assurances of accuracy, legitimacy, and diversity at scale, are an increasingly scarce resource—one that companies are holding close. As firms reach the limits of publicly available data (especially as more companies begin to create walls around previously open data), proprietary data will play an increasingly important role in training AI—meaning that Big Tech firms with their deep data feedback loops across their ecosystems have almost limitless access to valuable, diverse types of data.141See AI Now Institute, “Toxic Competition”; Open Markets, (<)em(>)Stopping Big Tech from Becoming Big AI: A Roadmap for Using competition Policy to Keep Artificial Intelligence Open for All(<)/em(>), October 17, 2024, (<)a href='https://www.openmarketsinstitute.org/publications/report-stopping-big-tech-big-ai-roadmap'(>)https://www.openmarketsinstitute.org/publications/report-stopping-big-tech-big-ai-roadmap(<)/a(>); and Tejas N. Narechania and Ganesh Sitaraman, “An Antimonopoly Approach to Governing Artificial Intelligence,” (<)em(>)Yale Law and Policy Review(<)/em(>) 43, no. 1 (Fall 2024), (<)a href='https://yalelawandpolicy.org/antimonopoly-approach-governing-artificial-intelligence'(>)https://yalelawandpolicy.org/antimonopoly-approach-governing-artificial-intelligence(<)/a(>). As high-quality data resources run out for non-Big Tech players, they will look to synthetic data—meaning either datasets containing data generated by other models, or AI-generated content (slop) that exists all over the internet.142Cade Metz et al., “How Tech Giants Cut Corners to Harvest Data for A.I.,” (<)em(>)New York Times(<)/em(>), April 6, 2024, (<)a href='https://www.nytimes.com/2024/04/06/technology/tech-giants-harvest-data-artificial-intelligence.html'(>)https://www.nytimes.com/2024/04/06/technology/tech-giants-harvest-data-artificial-intelligence.html(<)/a(>). While this may lower access to data in the short term, we should wonder what this will say about the quality of new models repeatedly trained on synthetic data of increasingly lower quality—while those incumbents with unlimited access to new data retain their data advantage. 

There’s a need to double down on advocating for data minimization to prevent Big Tech from using AI as its catch-all justification to use and combine data across contexts and store it forever, and limit aggressive strategies for acquiring companies and datasets to shore up their data advantages in high-value sectors or shut out competitors.143Amba Kak, Executive Director, AI Now Institute, (<)em(>)Safeguarding Data and Innovation: Building the Foundation for the Use of Artificial Intelligence, Before U.S. House Committee on Energy, Commerce, Subcommittee on Innovation, Data, and Commerce(<)/em(>), 118th Cong., 2023. Recent revelations that US DOGE’s operations, motivated by Elon Musk, may be feeding sensitive government data (including Medicare, Social Security payments, and taxpayer data) to train AI models144Charles Rollet and Zack Whittaker, “Elon Musk Staffer Created a DOGE AI Assistant for Making Government ‘Less Dumb,’” (<)em(>)TechCrunch(<)/em(>), February 18, 2025, (<)a href='https://techcrunch.com/2025/02/18/elon-musk-staffer-created-a-doge-ai-assistant-for-making-government-less-dumb'(>)https://techcrunch.com/2025/02/18/elon-musk-staffer-created-a-doge-ai-assistant-for-making-government-less-dumb(<)/a(>). for his company xAI brought these twin concerns into sharp focus. Government data, especially that obtained from and about the public, is an extremely valuable data asset for training AI. High-quality, vetted, and structured, this data covers a range of high-value sectors including healthcare, finance, and federal contracting. Amid frenzied AI competition, DOGE’s vast access to federal data could give the company an unfair advantage over peer AI companies—as well as create a clear conflict of interest with regard to Musk’s own involvement in spearheading DOGE. It’s also extremely sensitive data in terms of its ability to be used against the best interests of individuals and groups, without their control. This combination of privacy and competition risks will be on the rise, and it will be important to thread these concerns together in policy and advocacy fights, as well as to continue tracking145Kak and West, (<)em(>)2023 Landscape: Confronting Tech Power(<)/em(>). “data mergers” in direct and indirect forms.

Tech May Be Big, But It Isn’t a Monolith

Achieving legislative victories has been particularly challenging given an overall environment of enthusiasm for AI and fear that any intervention would hinder future prospects for innovation. Emerging points of fracture within the industry offer some points of potential leverage to achieve policy goals. For example:

We Need to Reclaim the Positive Agenda for Public-Centered Innovation, and AI Shouldn’t Be at the Center of It

Through 2024 we saw rapid momentum around the idea that AI development needs to be guided by non-market imperatives and that public and philanthropic capital can be put to the task of creating an innovation ecosystem outside of industry.154See Alex Pascal and Ganesh Sitaraman, (<)em(>)The National Security Case for Public AI(<)/em(>), Vanderbilt Policy Accelerator (VPA), 2024, (<)a href='https://cdn.vanderbilt.edu/vu-URL/wp-content/uploads/sites/412/2024/09/27201409/VPA-Paper-National-Security-Case-for-AI.pdf'(>)https://cdn.vanderbilt.edu/vu-URL/wp-content/uploads/sites/412/2024/09/27201409/VPA-Paper-National-Security-Case-for-AI.pdf(<)/a(>); Public AI Network, (<)em(>)Public AI: Infrastructure forthe Common Good(<)/em(>), August 10 2024, (<)a href='https://drive.google.com/file/d/1bcCPdRHyUGFB23--6wn4j1f9mRgdG2QF/view'(>)https://drive.google.com/file/d/1bcCPdRHyUGFB23–6wn4j1f9mRgdG2QF/view(<)/a(>); Eleanor Shearer et al., “The Role of Public Compute,” Ada Lovelace Institute (blog), April 24, 2024; Eleanor Shearer, “Take Back the Future: Democratising Power Over AI,” Common Wealth, September 11, 2024; Lawrence Lessig, “Why This is the Moment for Public AI,” keynote address, Building A More Public AI Ecosystem, YouTube, Aug 29, 2024, (<)a href='https://youtu.be/uzQDeYV1RJQ?si=nGCQFaprfZVjeFIn'(>)https://youtu.be/uzQDeYV1RJQ?si=nGCQFaprfZVjeFIn(<)/a(>); Bruce Schneier and Nathan Sanders, “Build AI by the People, for the People,” (<)em(>)Foreign Policy(<)/em(>), June 12, 2023, (<)a href='https://foreignpolicy.com/2023/06/12/ai-regulation-technology-us-china-eu-governance'(>)https://foreignpolicy.com/2023/06/12/ai-regulation-technology-us-china-eu-governance(<)/a(>); Nick Marda et al., “Creating a Public Counterpoint for AI,” Mozilla, (<)em(>)Distilled(<)/em(>) (blog),  September 30, 2024, (<)a href='https://blog.mozilla.org/en/mozilla/ai/public-ai-counterpoint'(>)https://blog.mozilla.org/en/mozilla/ai/public-ai-counterpoint.(<)/a(>) This includes a wide spectrum of proposals from think tanks, philanthropies, and governments ranging from introduction of public options to compete with private industry,155Narechania and Sitaraman, “An Antimonopoly Approach to Governing Artificial Intelligence.” to public-private arrangements aimed at increasing access to the inputs needed to build AI.156National Science Foundation, “National Artificial Intelligence Research Resource Pilot,” accessed May 3, 2025, (<)a href='https://www.nsf.gov/focus-areas/artificial-intelligence/nairr'(>)https://www.nsf.gov/focus-areas/artificial-intelligence/nairr(<)/a(>). Much of this activity was captured under the umbrella of “public AI,” while some of the efforts spearheaded by government have since been termed “sovereign AI,”157Bryce Elder, “The Next Wave of AI Hype Will Be Geopolitical. You’re Paying,” (<)em(>)Financial Times(<)/em(>), May 29, 2024, (<)a href='https://www.ft.com/content/a60c3c7b-1c48-485d-adb7-5bc2b7b1b650'(>)https://www.ft.com/content/a60c3c7b-1c48-485d-adb7-5bc2b7b1b650(<)/a(>). and more explicitly oriented around national security and competitiveness goals.158AI Now Institute, (<)em(>)AI Nationalism(s): Global Industrial Policy Approaches to AI(<)/em(>), March 12, 2024, (<)a href='https://ainowinstitute.org/publications/research/ai-nationalisms-global-industrial-policy-approaches-to-ai'(>)https://ainowinstitute.org/publications/research/ai-nationalisms-global-industrial-policy-approaches-to-ai(<)/a(>).

While this discourse drives home the realization that the current market-driven trajectory of AI is an impoverished one for the public, the alternative solution space as it stands has also tended to lack specificity, vision, and—most fundamentally—fails to follow through on the challenge to concentrated power, either accepting or evading the reality that building on the current large-scale AI paradigm will inevitably reinforce the power of the largest tech firms. For these reasons, it has been ripe for capture. 

Rejecting the current paradigm of large-scale AI is necessary to fight the information and power asymmetries vested in AI. This is the quiet part that we must say out loud. It’s the reality we must confront if we are to muster the will and the creativity to shape the situation differently.

The trajectory of the US National AI Research Resource (NAIRR) holds important lessons.

This is not to say that NAIRR hasn’t provided public value—the pilot matched data and compute resources to over three hundred university researchers and others who may not have otherwise had this access—potentially spurring socially beneficial projects.164NAIRR Pilot, “Resource Allocations,” accessed May 3, 2025, (<)a href='https://nairrpilot.org/projects/awarded'(>)https://nairrpilot.org/projects/awarded(<)/a(>). Yet there is a glaring missed opportunity: NAIRR could have spurred a widely deliberative and ambitious project that defined and fostered the conditions for a public-interest-oriented AI. Rather than building on the “bigger-is-better” trajectory in which all roads inevitably lead to doubling down on Big Tech infrastructure dependencies (fueling resources back to the same companies) or relying on a large, centralized state apparatus to shape the alternative vision, there was a moment for opening up space (and funding for) alternative research trajectories: conceptualizing a sustainable AI supply chain, investing in modes of data stewardship that are accountable to the communities technology is designed to serve, and widening the aperture beyond tech termed as “AI” to bring resources toward other technological gaps that affect the public (such as secure communications infrastructure).165Joana Varon et al., “AI Commons: Nourishing Alternatives to Big Tech Monoculture,” One Project, 2024, (<)a href='https://oneproject.org/ai-commons/'(>)https://oneproject.org/ai-commons(<)/a(>).

Despite current limitations, public AI is still a fertile opportunity space to both push a conversation around alternative trajectories for AI that are more structurally aligned with the public interest, as well as ensure that any public funding in this space is conditioned on public-interest goals. On the government side, while US NAIRR is on shaky footing, states are considering investments in public AI resources across the country. A California bill recently revived a proposal for a publicly owned and operated computer cluster housed in the University of California system called “CalCompute.”166Scott Wiener, “Senator Wiener Introduces Legislation to Protect AI Whistleblowers & Boost Responsible AI Development,” press release, February 28, 2025, (<)a href='https://sd11.senate.ca.gov/news/senator-wiener-introduces-legislation-protect-ai-whistleblowers-boost-responsible-ai'(>)https://sd11.senate.ca.gov/news/senator-wiener-introduces-legislation-protect-ai-whistleblowers-boost-responsible-ai(<)/a(>); Chase DiFeliciantonio, “Wiener Launches Bill to Require Testing AI Tools for Safety. Here’s What That Means,” (<)em(>)San Francisco Chronicle(<)/em(>), February 8, 2024, (<)a href='https://www.sfchronicle.com/tech/article/wiener-launches-bill-require-safety-testing-ai-18656225.php'(>)https://www.sfchronicle.com/tech/article/wiener-launches-bill-require-safety-testing-ai-18656225.php(<)/a(>). New York established an initiative called Empire AI to build public cloud infrastructure across seven New York state research institutions, bringing together over $400 million in public and private funds.167“Governor Hochul Unveils Fifth Proposal of 2024 State of the State: Empire AI Consortium to Make New York the National Leader in AI Research and Innovation,” New York State Governor, January 8, 2024, (<)a href='https://www.governor.ny.gov/news/governor-hochul-unveils-fifth-proposal-2024-state-state-empire-ai-consortium-make-new-york'(>)https://www.governor.ny.gov/news/governor-hochul-unveils-fifth-proposal-2024-state-state-empire-ai-consortium-make-new-york(<)/a(>). Both of these initiatives create important sites for advocacy to ensure their resources support communities’ needs and do not become a vehicle to further enrich Big Tech resources. 

Key Directions Forward for Public AI Efforts: 

Back: Chapter 3: Consulting the Record: AI Consistently Fails the Public