AI Now Institute

Theory of Change

AI Now's theory of change is built on a single core diagnosis: AI is not an independent technological phenomenon but an extension of corporate power concentration. Their causal chain:

Document harms from AI systems deployed in hiring, policing, healthcare, and other domains
Build public narratives that reframe AI as a corporate power problem, not a technical inevitability
Place staff in regulatory roles (three AI Now staffers served as FTC senior advisors in 2021-2022)
Drive legislation through congressional testimony, policy reports, and coalition partnerships
Constrain corporate power via antitrust enforcement, data privacy law, and algorithmic accountability

In their own words: "AI Now develops policy strategies to redirect away from the current trajectory: unbridled commercial surveillance, consolidation of power in very few companies, and a lack of public accountability."

Their 2025 report "Artificial Power" sharpens this to a specific bet: the AI bubble will burst because the technology lacks a viable business model. "Anthropic burned through $5.6 billion... OpenAI lost $5 billion... No profit-making use cases exist yet, or are even on the horizon."

They explicitly reject existential risk framing. Whittaker calls AGI fears "ghost stories" that serve as "advertisements for a technology that only a handful of companies have." Kak frames data privacy regulation as AI regulation: "If there's any single point I want to make today: it's that now is the moment where passing such a law matters most: before the trajectory has been set."

What They Do

Flagship outputs are annual Landscape Reports -- substantive (20K+ words), well-sourced policy documents. The 2023 "Confronting Tech Power" report was described by Vox as "a realistic roadmap for getting AI companies in check." The 2025 "Artificial Power" report argues AI is a bubble and proposes building alternatives to Big Tech's vision.

Direct policy engagement: Three staff served as FTC senior advisors on AI under Lina Khan (2021-2022). Amba Kak has testified before Congress three times (2023, 2024, 2025) and spoke before the UN General Assembly on AI governance (2025). Kak was one of only three civil society voices at the UK AI Safety Summit.

Research scope expanding: Originally focused on algorithmic bias, AI Now now covers compute and energy (nuclear regulation impacts), labor, safety/security (autonomous weapons evaluation), and AI governance. The 2025 "Fission for Algorithms" report analyzes how AI data center energy demands threaten nuclear safety standards.

Local organizing: Recent expansion into data center policy (North Star Data Center Policy Toolkit, December 2025) and community partnerships. Staff testified at Philadelphia City Council on AI issues.

Key People

Amba Kak (Co-Executive Director): The primary public policy face. Indian-origin, Oxford-educated, previously Mozilla policy advisor and FTC senior adviser on AI. Board member of Signal Foundation. TIME 100 Most Influential in AI 2024.

Meredith Whittaker (Chief Advisor, Co-Founder): The intellectual anchor but split between AI Now and Signal Foundation (where she is president). 13+ years at Google, co-organized the 2018 Google Walkout (20K employees), left citing retaliation. Her worldview -- that AI is a product of surveillance capitalism, not independent innovation -- defines AI Now's approach.

Heidy Khlaaf (Chief AI Scientist): Bridges traditional safety engineering and AI. PhD from UCL, previously led safety evaluation of OpenAI's Codex (methodology adopted industry-wide). MIT Technology Review Innovator Under 35. Focuses on autonomous weapons systems, finding AI accuracy rates as low as 25% in some military applications.

Team is ~21 named members, a lean think-tank structure with a small permanent staff augmented by fellows and visiting scholars. Co-founder Kate Crawford is no longer affiliated as of June 2024 with no public explanation.

Money and Incentives

Total budget: unknown. No 990 financial filings available despite 501(c)(3) status (EIN 33-3562042). This is the single biggest information gap.

Known funding:

Initial founding: $3.7M from MacArthur, Ford, Kapor Center, Open Society, Rockefeller (2017)
Additional $2M from MacArthur for national security/AI work (2022)
Part of Humanity AI $500M coalition (Oct 2025) -- $2M from MacArthur allocated to AI Now
Total known: ~$7.7M over ~8 years

Current funders (named, amounts unknown): Open Society Foundations, Ford Foundation, Mozilla Foundation, Omidyar Network, Luminate. Omidyar ecosystem provides at least 2 of 5 named funders.

Former funders (pre-2022): Microsoft Research, DeepMind, Ethics and Governance of AI Initiative, Pivotal Ventures (Melinda French Gates), Alfred P. Sloan Foundation, Robert Wood Johnson Foundation, Minderoo Foundation.

No corporate funding accepted since mid-2022 -- explicitly tied to examining tech companies' practices. No Coefficient Giving/Open Phil funding. Zero EA ecosystem funding. Funded entirely by mainstream progressive philanthropy.

Business model: Pure grant-funded nonprofit. No product revenue, consulting, or other income streams. Entirely dependent on foundation interest continuing.

Key incentive dynamics:

The no-corporate-funding policy is genuinely distinctive and removes the most obvious conflict of interest
Foundation funder concentration (especially Omidyar ecosystem) creates dependency risk
Critics (Daily Wire via InfluenceWatch) allege Omidyar-funded orgs coordinated FTC personnel placement -- the factual basis (Omidyar funds multiple groups with FTC connections) is documentable
Signal Foundation overlap: Whittaker (president + AI Now advisor), Kak (Signal board + AI Now co-ED). This interlocking directorate is unanalyzed in public record
No board of directors publicly identified, despite 501(c)(3) requirement

What Others Say

The strongest counterargument to AI Now's theory of change: By categorically dismissing existential risk -- calling it "ghost stories" and "religious fervor" rather than unlikely-but-possible -- AI Now may be making the same error they accuse x-risk advocates of: ignoring a class of risks because it doesn't affect the constituency they prioritize. The Brookings Institution notes "both camps have valid points, and the optimal policy response likely addresses both near-term and long-term risks." An academic paper (arxiv:2512.10058) documents that the divide between AI ethics and AI safety communities means neither camp fully engages with the other's best arguments.

The strongest counterargument to AI Now's empirical claims: Whittaker's dismissal of AI capabilities -- "It is not accurate. It is not trustworthy for any domain where facts actually matter" -- is testable and increasingly strained. The 2025 report's bet that the AI bubble will burst is falsifiable. If frontier AI capabilities continue improving and finding commercial applications, AI Now's core narrative weakens significantly.

InfluenceWatch: Characterizes AI Now as part of an Omidyar-funded network placing personnel in regulatory bodies. The factual claim about coordinated philanthropic influence is worth noting.

Mainstream recognition: Vox calls their 2023 report "refreshingly pragmatic and actionable." Kak named TIME 100 Most Influential in AI 2024. Khlaaf named MIT Technology Review Innovator Under 35. These indicate strong credibility in policy circles.

Notable silence: The x-risk/AI safety community has not produced targeted critiques of AI Now specifically. The debate exists at the community level (AI ethics vs AI safety) but not as direct organizational engagement.

What's Absent

Financial transparency: No 990 filings, no public budget, no known dollar amounts per funder. For an org advocating accountability, this is ironic.
Board disclosure: Board members not publicly identified despite 501(c)(3) status.
Crawford departure explanation: Co-founder leaving without public statement.
Concrete policy attribution: No documented cases where a specific policy changed because of AI Now's work (vs. the broader movement).
Engagement with technical AI safety: No position on alignment research, interpretability, or whether technical approaches to AI risk are valuable.
Success metrics: No public self-evaluation of whether their theory of change is working.
Any positive vision for AI: All published work focuses on harms, never on ways AI might address the very problems (inequality, surveillance, labor exploitation) they care about. Their explicit critique of "AI for Good" framing suggests this is intentional.