FATHOM

OLYMPUS opened an institutional file on Fathom because it is the organization that wrote the template for a state-run market in AI evaluators. This is not a psychometric profile. It is a statute, a partner list and a filing. The finding is the shape of the arrangement: a philanthropically funded nonprofit sponsored the California law under which outside testers of AI systems are designated by the state, lists among its partners several organizations that test, audit or insure AI systems, funds one of them, shares advisers with another, and has a president who sits on the board of a third. Every one of those ties is on a page the parties published. None is evidence of what any designation committee will do, and the statute has not yet designated anyone.

Institutional Archetype

THE ARCHITECT OF THE MARKET IN GRADERS — Fathom describes itself as “an independent nonprofit whose mission is to build a global architecture that helps society navigate the transition to a world with AI by fostering trust, safety, and innovation” (Fathom, About). Its product is one design, the independent verification organization, or IVO: a body of outside experts that the government authorizes, that tests AI systems against government-set outcomes, and that loses its license if it stops being independent of the companies it tests. Its own site calls the resulting arrangement an “Independent Oversight Marketplace” and says it is “a third-way approach to AI governance” (Fathom, Independent Oversight Marketplace for AI). Fathom did not test a model, grade a lab or write a benchmark. It wrote the rule that decides who may grade. This profile scores the rule-writer’s position, not any person or any evaluator it works with.

Mandate & Origin

• Launch. Fathom, “an organization mapping the ambitions of a broad and diverse society to the specific steps required to move toward its vision for a world with AI,” “announced today that it is launching with $25M dedicated to its efforts” on 23 July 2024 (Fathom, launch release). ProPublica’s record for Fathom AI Inc lists “Tax exemption issued: Jan. 2024” and EIN 93-4840479, and Fathom’s own footer prints “Fathom is a 501(c)(3) nonprofit FEIN 93-4840479” (ProPublica, Fathom AI Inc; Fathom, About).
• Leadership. The team page lists “Andrew Freedman Co-Founder, Chief Executive Officer” and “Bri Treece Co-Founder, President” (Fathom, About); see Bri Treece. The FY2024 Form 990 as ProPublica summarizes it lists “Andrew Freedman (Co-founder & Chief Stategy Officer)” at $156,500, Treece as Co-founder and Chief Impact Officer at $87,000 and “Blake Pierson (Co-founder & Ceo)” at $30,840. Fathom’s team page does not list Pierson (ProPublica, Fathom AI Inc).
• The vehicle. Fathom “has developed and championed the IVO model at the heart of SB 813” and says it “sponsored SB 813 to begin building the next layer of accountability: the bill directs the Government Operations Agency to designate qualified IVOs” (Fathom, 31 August 2026). The Senate Judiciary analysis of 29 April 2025 and the Assembly Privacy analysis of 1 July 2026 both name Fathom as sponsor (Senate Judiciary analysis; Assembly Privacy analysis).
• The bill changed shape on the way. The March 2025 version created “an immunity shield against personal injury and property damage caused by AI models and applications that are certified at the time of the injuries by a private entity designated by the AG, called an MRO,” and the analysis says “This bill is sponsored by Fathom, an entity with hopes of becoming an MRO” (Senate Judiciary analysis). By June 2026 it created a seven-member standards commission with a two-tier standards scheme and “criteria for designating qualified auditors as independent verification organizations” (Assembly Privacy analysis). As chaptered on 9 September 2026 it contains no commission and no liability shield: the Government Operations Agency sets application requirements and criteria, working groups advise it, and an audit performed to a published standard “is relevant to, but not conclusive of, the action” (leginfo, SB 813 as chaptered).

Funding & Backers

• What Fathom says. “Fathom is a 501(c)(3) organization funded by philanthropists. We do not take donations from corporations, including frontier labs and the FAANG companies, or foreign entities associated with countries of concern” (Fathom, 31 August 2026 release footer). Its IVO FAQ, in answer to “Who funds you?”, says the gifts are unrestricted and that its philanthropic partners have no influence over its strategy and approach (Fathom, IVO FAQ).
• What the filing shows. FY2024 revenue $12,128,490, of which “Contributions $12,127,300 100.0%”; expenses $5,832,833; net assets $6,295,657 (ProPublica, Fathom AI Inc). No donor is named in the summary; Schedule B donor identities are not part of it.
• Donors, named or unnamed. None is named on any Fathom page read, in the launch release or in the filing summary. Fathom’s team page lists a Chief of Philanthropic Partnerships and a Director of Philanthropic Partnerships and names no funder (Fathom, About). The one documented flow from Fathom to a third party is outbound: AVERI’s about page lists Fathom among its funders (AVERI, About), under the Alliance Map below.
• Board. The FY2024 Form 990 summary lists “Max Henderson (Board Member)” and “Kellie Liket (Board Member),” each at $0 (ProPublica, Fathom AI Inc). The FY2024 Form 990 summary for Halcyon Futures Inc (EIN 93-3537739) lists “Michael Mccormick (Chief Executive Officer),” “Max Henderson (Secretary)” and “Kellie Liket (Director)” (ProPublica, Halcyon Futures Inc). AVERI’s team page lists “Max Henderson” under Board (“Max is a board member at AVERI and the founder of Ergo Impact”) and Mike McCormick as board member of AVERI and “the founder and CEO of Halcyon Futures” (AVERI, Team). Halcyon Futures Inc’s FY2024 Schedule R lists Fathom AI, Ergo Impact Inc and Sentinel Bio Inc as related tax-exempt organizations (ProPublica, Halcyon Schedule R); Ergo Impact’s own page lists “Max Henderson CO-FOUNDER, CEO” and “Kellie Liket CO-FOUNDER, PRESIDENT” (Ergo Impact), and Fathom’s FY2024 return lists both as Board Member. See Halcyon Futures. AVERI’s page ties McCormick to Halcyon explicitly; Halcyon is also on AVERI’s funder list.

• Related, on one side only. Fathom’s own FY2024 return answers No to the related-organizations question (Part IV, line 34), as Sentinel Bio’s does, while Halcyon Futures Inc’s FY2024 Schedule R lists both as related (ProPublica, Fathom AI Inc; ProPublica, Sentinel Bio FY2024 Form 990; ProPublica, Halcyon Schedule R).

Institutional Voice & Intent

Fathom speaks in the register of a convener with a product: bipartisan, market-minded, and impatient with fixed rules. Its IVO site says “Direct regulations and licensing regimes, the traditional modes of governance, are too static and inflexible for a fast-evolving, all-encompassing technology” and that “IVOs turn safety standards into a competitive advantage” (Fathom, Independent Oversight Marketplace for AI). Its CEO’s line on the day SB 813 passed: “The power of the independent verification model is that it lets independent, qualified experts assess the risk of AI capabilities, and it keeps pace with the technology instead of freezing a single test into law” (Fathom, 31 August 2026).

Stated intent: Build a state-authorized market of independent testers, then replicate it in other states and in Congress. Observed intent: The same, on the legislative record: a bill in Sacramento, a companion auditor registry, an executive order that accelerates the criteria deadline, and a federal bill that licenses IVOs. Gap: None on the direction. The open question is structural and the record does not settle it: the organization that designed the licensing rule lists as partners, funds, advises and shares board members with organizations in the class the rule licenses. Each tie is disclosed on the parties’ own pages; none is shown to have shaped a designation, because there has been none.

Interest. Fathom argues four positions. For each: what adoption would pay, as documented; its best case in its own words; who else gains from the same rule; and the documented opposition in its own words.

• A state-designated market of independent verifiers (SB 813). What it pays: the rule creates a class, the IVO, defined as “an AI auditor that is designated by the agency as having demonstrated expertise in assessing the risks posed by an AI system or model,” where an AI auditor is “a person, partnership, academic institution, nonprofit, or corporation that conducts a covered audit on behalf of a third party,” and an IVO “may accept payment from a party being assessed at reasonable market rates” so long as payment is not conditioned on the result (leginfo, SB 813 as chaptered). The companion AB 1405, Chapter 178, creates an AI Auditor Registry by 1 January 2029, with annual registration fees, and from that date “prohibit[s] an unregistered person from offering, selling, or conducting a covered AI audit” (leginfo, AB 1405 as chaptered). Fathom’s team page lists a Vice President of AI Assurance Provider Services, and its partner list names organizations that test, audit or insure AI systems (Fathom, About; Fathom, Partnerships). No fee, designation or contract to Fathom or to any named partner is documented. Best case, Fathom’s: “It gives the public a way to trust AI and gives companies a credible way to show their system safeguards work as intended,” and “determining whether these AI models are safe still largely relies on companies to assess their own systems and report their own results” (Fathom, 31 August 2026). On capture: “IVOs must perpetually demonstrate independence from industry in order to secure a state license” and an IVO that sets unattainable standards loses business to a competitor, in its words (Fathom, IVO FAQ). Same rule helps: the evaluators that would apply; OpenAI, which endorses “SB 813 on overall infrastructure for independent safety assessments” (OpenAI, AI policy window); Omidyar Network, which writes of “a flexible, market-based and future-oriented approach” (Assembly Privacy analysis). Documented opposition, in its own words: TechNet, on the June 2026 version, “a state-specific definition of “substantially exceeds compliance” adds to a growing patchwork rather than reducing it”; the California Chamber of Commerce, “No seven individuals can reasonably be expected to possess the collective expertise necessary to evaluate technical standards, auditing methodologies, and risk-assessment frameworks across all of these domains simultaneously”; and, on the March 2025 version, TechNet: “Rather than establishing bespoke organizations and frameworks from scratch, we encourage California to build on the work of established, globally respected standards-setting bodies” (Assembly Privacy analysis; Senate Judiciary analysis).
• Legal protection as the incentive to be verified. What it pays: Fathom’s IVO site says “A state could choose to heighten the incentive for companies to participate, potentially by authorizing the verifiers to confer legal certainty or tort protection,” and its account of its August 2025 exercise describes “the incentive of heightened legal protections on being verified by the IVO” (Fathom, IVO site; Fathom, tabletop exercise). The March 2025 bill gave certified models an immunity shield; amendments replaced it with a rebuttable presumption and the chaptered text keeps only “relevant to, but not conclusive of.” The federal FRONTIER Act, H.R. 9925, goes the other way for the verifier: “An IVO shall be immune from suit and liability under Federal and State law with respect to all claims for loss caused by” the materialization of a catastrophic risk of a model it assessed, with an exception for willful misconduct causing death or serious physical injury (Congress.gov, H.R. 9925 text). Fathom’s statement on the bill does not mention the immunity clause; Fathom’s position on it is not stated in any page read. Best case, Fathom’s: the incentive is what makes a voluntary system voluntary for companies to join. Opposition, in its own words, the Consumer Attorneys of California on the 2025 bill: “Granting immunity erases the very purpose of civil justice” (Senate Judiciary analysis).
• Federal licensing and a mandatory verifier for the largest developers (FRONTIER Act). What it pays: H.R. 9925, introduced 23 July 2026, provides that “a very large frontier developer shall retain an independent verification organization that has been licensed,” so adoption turns a voluntary market into one with a statutory buyer (Congress.gov, H.R. 9925; text). Best case, Fathom’s: “What will work is a competitive market of independent verifiers who are accountable for real-world outcomes and who the government can actually count on,” and the bill “licenses competing verification organizations, allows them to sharpen their methods, and gives the government the power to revoke a license when a verifier’s work does not hold up in the field” (Fathom, FRONTIER Act statement). Same rule helps: the licensed evaluators, and the labs, which would hold a licensed verifier’s assessment. No opposition statement on the bill was read.
• More verification, inside the labs (the Governor’s order). What it pays: Executive Order N-9-26 asks for recommendations by 16 November 2026 on “Requiring that all large frontier developers embed designated independent verification organizations onsite in their labs to conduct periodic audits and evaluations” and on a “kill switch” whose efficacy is “verified on an ongoing basis by an independent verification organization”; these are recommendations, not requirements (Executive Order N-9-26). Fathom says the order “looks at other places where independent checks could make AI safer” (Fathom, statement on executive order). Same rule helps: Anthropic, which proposes that “Each frontier AI company commits to giving ongoing, employee-like access to a team of embedded third-party evaluators (such as METR)” (Dario Amodei); see METR and AI Evaluator Forum.

Interest is a documented fact about each position, not a claim about motive, and not a verdict that the argument is wrong. Each best case is stated at the strength its author states it.

Position in the Apparatus

Fathom sits upstream of the evaluators. METR, Transluce, FAR.AI, Apollo Research and the members of the AI Evaluator Forum test models; the Forum writes the conditions under which a test counts as independent. Fathom proposes who may be designated to test and under what independence criteria, and the statute it sponsored assigns that work to a state agency. The nearest peers in the map are CAIS, whose Action Fund sponsored SB 1047 and whose former Chief Operating Officer is Fathom’s President, and GovAI, which supplies research and people rather than a bill. The Frontier Model Forum is the labs’ own table and says it does not lobby. Fathom is the one organization in the group that wrote a statute and counts evaluators among its partners.

Which evaluators sit in or near the market

The statute names no entity, and no one has been designated. The table lists organizations that, on the parties’ own pages, have a documented tie to Fathom, to SB 813 or to both, and what the tie is.

Who testified and who opposed

• Senate Judiciary, 29 April 2025 (the March 2025 version): the committee voted 10-0 to pass the bill as amended. The analysis lists “Fathom (sponsor)” and “21 individuals” in support and seven organizations in opposition: the California Chamber of Commerce, the California Initiative for Technology & Democracy, Chamber of Progress, the Children’s Advocacy Institute, Consumer Attorneys of California, Tech Equity Action and TechNet. A coalition of CITED, the Children’s Advocacy Institute and Tech Equity wrote of the 2025 bill: “It is not clear whether there are any entities currently in the United States other than perhaps the sponsor of this legislation who might be able to comply with the current MRO qualifications in the measure” (Senate Judiciary analysis; leginfo, history). CalMatters’ Digital Democracy lists Andrew Freedman as testifying in support at that hearing, as reported (Digital Democracy, Fathom).
• Assembly Privacy and Consumer Protection, 1 July 2026 (the June 2026 version): 11-2. The analysis’s registered support: Fathom (sponsor), Children Now, Eticas, the Little Hoover Commission, Meridian Governance INC., Mothers Against Media Addiction, Oakland Privacy, Omidyar Network, The Alliance for Secure AI, Transparency Coalition.ai and four academics; support if amended: Encode AI Corporation and Secure AI Project; opposed: TechNet and the Valley Industry and Commerce Association; oppose unless amended: the California Chamber of Commerce (Assembly Privacy analysis; leginfo, history). As reported by Digital Democracy’s transcript of the hearing, Nichole Rocha spoke for Fathom and Jai Jaisimha for the Transparency Coalition; Robert Boykin spoke for TechNet and Ronak Daylami for the California Chamber of Commerce; the Electronic Frontier Foundation, which had opposed the earlier text, said it was neutral on the bill as in print, and Oakland Privacy and Tech Equity said they had dropped opposition (Digital Democracy, hearing of 1 July 2026).
• The votes, from the Legislature’s own record. Senate Appropriations 5-2 (22 January 2026); Senate floor 31-7 (27 January 2026), the noes Alvarado-Gil, Choi, Dahle, Grove, Jones, Seyarto and Strickland; Assembly Appropriations 11-3 (13 August 2026); Assembly floor 67-6 (30 August 2026), the noes DeMaio, Johnson, Lackey, Sanchez, Ta and Tangipa; Senate concurrence, “Ayes 37. Noes 0.” per the history line (30 August 2026); approved by the Governor and chaptered as Chapter 179, Statutes of 2026, on 9 September 2026 (leginfo, history; leginfo, votes). The author is Senator Jerry McNerney; Assemblymember Rebecca Bauer-Kahan carried it on the Assembly floor and is the author of the companion AB 1405 (Chapter 178) (Governor, 9 September 2026).

Actions & Leadership Choices

• The Governor’s order. On 18 September 2026 the Governor signed Executive Order N-9-26: “No later than May 1, 2027, the Government Operations Agency shall complete the requirements of Section 8898.1 of the Government Code,” against the statute’s 1 January 2028; recommendations on embedded evaluators, independent verification of the safety filings labs already submit, and a kill switch are due 16 November 2026 (Executive Order N-9-26). Fathom’s statement of the same day calls the order progress that “gets California there faster” (Fathom, statement on executive order).
• The Governor’s expert group. On 23 September 2026 Fathom announced that “key architect of the IVO (Independent Verification Organization) framework and Fathom senior advisor, Dr. Gillian Hadfield, was named to Governor Gavin Newsom’s group of AI experts”; the Governor’s office lists Hadfield among four named experts who advise on the order’s recommendations (Fathom, 23 September 2026; Governor, 23 September 2026). Fathom’s team page lists Hadfield as an affiliate member, and AVERI’s team page lists her as an advisor (Fathom, About; AVERI, Team). Freedman says “Two years ago we teamed up, and Fathom has proudly worked alongside Dr. Hadfield to bring that idea to life.”
• What the statute and the sponsor’s description say. Fathom’s statement says “A separate, expert-staffed body, not the IVOs or the labs themselves, sets the standards IVOs are assessed against.” The chaptered text assigns the criteria to the Government Operations Agency and requires working groups that “shall include, but not be limited to, engineers from AI companies that are competitors and AI safety experts”; it requires an IVO to report annual changes to “governance policies or sources of funding relevant to the IVOs conflicts of interest or independence” (Fathom, statement on executive order; leginfo, SB 813 as chaptered).
• Federal. Fathom welcomed H.R. 9925, sponsored by Rep. Obernolte and introduced 23 July 2026, and asked for “a fuller range of tools to act upstream” (Fathom, FRONTIER Act statement; Congress.gov).
• Prototyping the market. Fathom’s August 2025 exercise had participants “play the role of an IVO” and run a verification of a model against outcome metrics, using Transluce’s Docent for the assessment; it reports that the IVO model was “seen to be technically feasible” by participants (Fathom, tabletop exercise). Its July 2026 piece “Who Evaluates the Evaluators?” says “Many evaluators also hold commercial relationships with the very companies whose systems they might evaluate” and quotes Chowdhury: “Being a purely independent evaluator is nearly impossible, because it’s just not a fully built-out ecosystem yet” (Fathom, Who Evaluates the Evaluators?).
• Convening. Fathom lists The Ashby Workshops as its annual invite-only convening (Fathom, Ashby Workshops 2026).

CONDUCT: WATCH — SPONSOR OF THE DESIGNATION RULE, WITH DISCLOSED TIES TO THE EVALUATORS IT CALLS PARTNERS. The work is public, the sponsorship is stated on the statute’s own committee analyses, and the funding policy is on the front page. What the structure shows: the organization that designed the rule for licensing evaluators funds one evaluator, shares two advisers with it, lists six more organizations as partners, has a president on the board of another, and published a piece on the field’s missing independence rules. It is a documented seam, not a documented conflict.

Alliance Map

The overlap is the finding. On one set of documents Fathom is at once the designer of a state licensing rule for AI evaluators, a funder of one evaluator, a partner of six more, an organization whose senior advisor (Hadfield) and President (Treece) both advise that funded evaluator while Treece also sits on FAR.AI’s board, a convener with the Foundation for American Innovation whose officer sits on AVERI’s board, and the sponsor of the law whose criteria an order of the Governor has just accelerated. Each tie is disclosed by the parties. The sources show disclosed seats, listings and funding, not coordination, and no designation has been made.

Reach Assessment

Institutional: Moderate and rising. The statute binds the state agency, not a developer: it says it does not require any developer to engage an IVO, and an IVO is paid at market rates by whoever hires it. What it changes is who the state will call qualified, by 1 May 2027 on the Governor’s order. The federal bill would make the same designer’s model a purchase obligation for the largest developers.

Memetic: High within the policy field. “IVO” is now the term of art in state bills, a federal bill and an executive order, and the OpenAI and Anthropic positions in the field’s debate use the vocabulary (OpenAI, AI policy window; Dario Amodei).

Civilizational: Low to moderate, structural. Fathom grades nothing and builds no model. The hazard is the seam: who writes the license standard for evaluators and who funds, advises and sits on the boards of those the license covers.

Score basis. Atk 7 is reach: Fathom sponsored a statute that the Legislature passed 67-6 in the Assembly and the Governor signed as Chapter 179, and an executive order of 18 September 2026 moved the IVO criteria deadline to 1 May 2027 (leginfo, history; Executive Order N-9-26). The points under 9: the statute does not “Require any person, partnership, or corporation that develops, deploys, or operates an AI system or model to engage an IVO,” it designates no one yet, and the federal bill is introduced, not law (leginfo, SB 813 as chaptered; Congress.gov). Def 7 is funding insulation: FY2024 contributions of $12,127,300 were 100.0 percent of revenue, and Fathom takes no donations from corporations or frontier labs (ProPublica, Fathom AI Inc; Fathom, 31 August 2026). The points under 8: no donor is named on any page read, and the money is spread over an unknown number of philanthropists, so the filing cannot show concentration or independence from any one of them. Hp 6 is entrenchment: two chaptered statutes and an order with dated deadlines outlast the organization, against which the statute is voluntary, designation criteria are not published, and Fathom itself has net assets of $6,295,657 and a two-year record (Executive Order N-9-26; ProPublica, Fathom AI Inc). Peers: CAIS at 9/7/6, hazard 83, holds more reach through a signatory list the labs’ chiefs sit on, the same insulation and the same entrenchment; the Frontier Model Forum at 9/6/7, hazard 79, has more reach through a side agreement six labs signed, one point less insulation because members pay its fees, and one more entrenchment through three-year fee commitments; GovAI at 7/8/7, hazard 76, shares the reach, with more insulation and more entrenchment; the AI Evaluator Forum at 5/6/4, hazard 50, has less of each because it has no legal form, no budget and no statute.

Sources: Fathom — About; Fathom — Partnerships; Fathom — Independent Oversight Marketplace for AI; Fathom — IVO FAQ; Fathom — launch release, 23 July 2024; Fathom — California Legislature passes SB 813; Fathom — statement on Governor Newsom’s executive order; Fathom — FRONTIER Act statement; Fathom — Gillian Hadfield appointed; Fathom — Behind the Scenes of an IVO; Fathom — Who Evaluates the Evaluators?; The Ashby Workshops 2026; leginfo — SB 813 text; leginfo — SB 813 history; leginfo — SB 813 votes; leginfo — AB 1405 text; Senate Judiciary Committee analysis of SB 813, 29 April 2025; Assembly Privacy and Consumer Protection analysis of SB 813, 1 July 2026; CalMatters Digital Democracy — hearing of 1 July 2026; CalMatters Digital Democracy — Fathom; Executive Order N-9-26; Governor — signs AI safeguards, 9 September 2026; Governor — announces experts, 23 September 2026; Congress.gov — H.R. 9925; Congress.gov — H.R. 9925 text; AVERI — About; AVERI — Team; Miles Brundage — The Launch of AVERI; FAR.AI — team; ProPublica — Fathom AI Inc; ProPublica — Halcyon Futures Inc; PIB — IndiaAI Mission, 12 January 2026; OpenAI — AI policy window; Dario Amodei — We Must Pace the Frontier.