Fathom is a US nonprofit working on AI policy, which describes itself as "an independent nonprofit advancing policies to ensure AI is safe, secure, and trusted" and states its mission as building "a global architecture that helps society navigate the transition to a world with AI by fostering trust, safety, and innovation" (Source: fathom.org). Its chief executive Andrew Freedman describes it as "an AI governance nonprofit that finds, builds, and scales policy and technical innovations designed for the AI century" (Source: transformernews.ai).
Fathom advocates for the independent verification organization (IVO) model — private bodies, accredited by government, that certify AI developers against safety outcomes the state sets. Gabriel Weil's July 2026 critique of the model identifies Fathom as the organization behind the model legislation carrying it (Source: Don't Let AI Developers Hire Their Own Referees (Weil, July 2026)). Instruments associated with the design include California SB 813 (AI Standards and Safety Commission), Connecticut's IVO pilot under Connecticut SB 5 — Broad AI law (frontier reporting + ADMT + AI companions + sandbox), and the federal FRONTIER Act; the structure and status of the first two are recorded as contested on Independent Verification Organizations (IVOs). Fathom published a statement on the FRONTIER Act's introduction describing it as "the first federal blueprint for independent AI verification" (Source: fathom.org).
Organization
Andrew Freedman is co-founder and chief executive officer; Bri Treece is co-founder and president (Source: fathom.org). Earlier coverage of Fathom's work on California SB 813 identifies Freedman as chief strategy officer, a title the organization's current staff listing no longer carries (Source: csis.org). Staff roles include a vice president of state policy and external affairs and a vice president of AI assurance provider services — the latter a function tied directly to building the verifier market the organization advocates for (Source: fathom.org).
Fathom lists a set of affiliate members drawn from AI governance research, national security and philosophy, including Gillian K. Hadfield, Yoshua Bengio, Michèle Flournoy of WestExec Advisors, Danielle Allen of the Harvard Kennedy School, Paul Scharre of CNAS, Robert Trager of the Oxford Martin AI Governance Initiative, Jeff Alstott of the RAND Corporation Center for Technology and Security Policy, Zach Graves of the Foundation for American Innovation (FAI), and Gregory C. Allen of Center for Strategic and International Studies (CSIS) (Source: fathom.org). Hadfield and Freedman have presented the IVO design jointly as a proposal to govern AI through private regulatory markets (Source: cognitiverevolution.ai).
Position on independent verification
Freedman set out the organization's argument in AI Frontiers on July 1, 2026, in response to the Supreme Court's decision in Trump v. Slaughter permitting at-will removal of independent agency heads. His claim is that the ruling strengthens rather than weakens the case for private verification: since "there is no such thing as an independent federal agency," AI governance should "stop trying to insulate the politics" and "insulate the facts instead" (Source: transformernews.ai).
The architecture he proposes separates two functions the field has treated as one. The technical task of measuring model capability and danger sits with CAISI inside the Commerce Department, which would "run and validate evaluations, build benchmarks, set measurement standards, hold secure channels for model access, and publish what it finds" without making binding decisions. Around it would grow "a competitive ecosystem of accredited private testers" working to thresholds set by Congress or through rulemaking and reporting to CAISI. Competition, in this account, "is not a market slogan but a mechanism that forces the science of measuring AI safety to keep pace with the capabilities it measures"; an IVO that loses its independence loses its accreditation. The political decisions — how safe is safe enough, what follows a shortfall, which violations draw penalties — remain with elected institutions (Source: transformernews.ai).
Freedman anticipates the constitutional objection that private bodies would exercise government power, arguing they would not: "The coercive choices stay with the government. The verification organizations only measure and assess, the way financial auditors and product-safety labs do." He acknowledges the design risks directly — "Competition can be abused. Standards can be written badly. Accreditation can curdle into a rubber stamp" — while treating them as tractable (Source: transformernews.ai).
Contested claims about enacted IVO instruments
Freedman states that "Connecticut and Virginia have already passed bills establishing independent verification for AI governance" (Source: transformernews.ai). This conflicts with the account in Gabriel Weil's July 29, 2026 essay, which describes Virginia as having directed a commission to study the model rather than enact it (Source: Don't Let AI Developers Hire Their Own Referees (Weil, July 2026)). The instrument in question is not Virginia HB 2094 (High-Risk AI Developer and Deployer Act, vetoed), which was vetoed in March 2025 and is a Colorado-style high-risk bill rather than an IVO measure. The two accounts have not been reconciled against the primary record, and the discrepancy is recorded on Independent Verification Organizations (IVOs) as differing accounts.
Freedman also writes that "the argument over whether independent verification is the right home for the science of AI governance is nearly settled," citing convergence by Anthropic, OpenAI and Google DeepMind (Source: transformernews.ai). Weil's essay is a direct counter: it argues that a model in which developers select and pay their own certifiers reproduces the conflict of interest that discredited credit-rating agencies after 2008 (Source: Don't Let AI Developers Hire Their Own Referees (Weil, July 2026)).
Open questions
- Which Virginia instrument Freedman refers to, and whether it establishes verification or only studies it.
- How accreditation would be designed to prevent the rubber-stamp failure Freedman names as a risk, given that developers would select and pay their verifiers.
- Whether Fathom's assurance-provider services function places it in the verifier market its policy work is helping to design, and what disclosure that would call for.
Relationships
- supports: Independent Verification Organizations (IVOs) — the organization is the model's principal advocate
- supports: California SB 813 (AI Standards and Safety Commission), Connecticut SB 5 — Broad AI law (frontier reporting + ADMT + AI companions + sandbox), Frontier Act / Great American AI Act (Obernolte–Trahan) — instruments carrying the IVO design it promotes
- contradicts: Don't Let AI Developers Hire Their Own Referees (Weil, July 2026) — Weil argues the developer-pays structure reproduces the credit-rating-agency conflict Fathom's design does not resolve
- depends-on: NIST CAISI (Center for AI Standards and Innovation) — the federal body Fathom's architecture puts at the technical center
- related: Trump v. Slaughter — the decision Freedman treats as the case for private verification
- related: Gillian K. Hadfield — affiliate member and source of the regulatory-markets scholarship behind the IVO design
- related: AI Governance (umbrella), AI Federalism