AVERI

OLYMPUS opened an institutional file on AVERI because it is the evaluator on the map that argues for the rule that would create the market in evaluators. The other seven members of the AI Evaluator Forum test models, write benchmarks or run leaderboards. AVERI mostly writes the case for requiring audits, the standards an audit would be held to and the quality-control body that would police the auditors, and runs pilots to test the pieces. This is not a psychometric profile. It is a funders list, a board page and a set of position papers. The finding is the shape of the arrangement: an audit advocate funded in part by the sponsor of California’s evaluator-designation law and by donors that also appear behind other evaluators, governed by a board that includes the chief executive of a grant fund on its funders list, who also sits on another evaluator’s board, and advised by that sponsor’s President and a senior advisor to it. Every tie is on a page the parties published. None is evidence that any funder shaped any finding, and the page-readable record on AVERI is short, because the organization is about fifteen months old.

Institutional Archetype

THE AUDITOR WHO WRITES THE AUDIT STANDARD — AVERI describes its problem in one sentence: “Safety and security claims about AI systems are largely self-reported by companies” (AVERI). Its mission is “to make frontier AI auditing effective and universal,” and it describes its method as “We are a standard-setter grounded in practice” (AVERI, About). It is a think tank, in its founder’s words, not an audit firm: “We’re not trying to be the auditors exactly” (Miles Brundage, The Launch of AVERI). What it produces is the vision paper, the pilot reports, a coauthored evaluator standard and bill endorsements. Alone among the Forum’s eight, it is organized around a requirement that does not yet exist in federal law: “Frontier AI audits aren’t yet required” (AVERI, About). This profile scores the position, not any person.

Mandate & Origin

• Founding. Brundage’s launch post of 15 January 2026 calls AVERI “the non-profit I cofounded last year” and says it “has officially launched” that day (Miles Brundage, The Launch of AVERI). AVERI’s team page says of board member Grace Werner, “Previously, Grace co-founded AVERI and served as Chief of Staff and lead U.S. policy” (AVERI, Team). Fortune, as AVERI’s own listing reprints it, reported: “Today Brundage formally announced the AI Verification and Evaluation Research Institute (AVERI), a new nonprofit aimed at pushing the idea that frontier AI models should be subject to external auditing” (AVERI, launch announcement). The launch coincided with the flagship paper, “Frontier AI Auditing: Toward Rigorous Third-Party Assessment of Safety and Security Practices at Leading AI Companies,” dated 15 January 2026 (AVERI, Frontier AI Auditing).
• Legal form. “The AI Verification and Evaluation Research Institute® (AVERI, which rhymes with bravery) is a 501(c)(3) US-based non-profit organization” (AVERI). ProPublica’s Nonprofit Explorer lists Ai Verification And Evaluation Research Institute, 601 13th St NW, Washington, DC, EIN 38-4354994, “Tax exemption issued: July 2025,” and “Nonprofit Explorer has no Form 990 data for Ai Verification And Evaluation Research Institute” (ProPublica Nonprofit Explorer).
• Leadership. Miles Brundage, Executive Director: “Previously, Miles led the Policy Research and AGI Readiness teams at OpenAI”; he is a Non-Resident Senior Fellow at the Institute for Progress, “an advisor to Epoch AI and the RAND Corporation” and “a member of the EU AI Act Scientific Panel.” The team page lists, beyond him, Carly Tryens (Executive Operations), Sean McGregor (Lead Research Engineer), Jacqueline Grimsley (Operations Director), Nat Purser (US Policy Director), George Balston (Policy and Advocacy Lead, who “leads the organization’s work on building demand for frontier AI auditing”) and Patricia Paskov (Standards Director, “also an Adjunct Researcher at RAND, a Research Affiliate at the Oxford Martin AI Governance Initiative”) (AVERI, Team).
• Board. Three members: Max Henderson (“Max is a board member at AVERI and the founder of Ergo Impact”), Mike McCormick (“the founder and CEO of Halcyon Futures”) and Grace Werner (“Vice President of Growth at the Foundation for American Innovation (FAI)”) (AVERI, Team).
• Advisors. David Duvenaud, A. Feder Cooper, Gillian Hadfield, Jose Hernandez-Orallo, Bri Treece and Jeff Wu. Treece is listed as “Co-founder and Chief Impact Officer of Fathom,” where “most recently, she served as the COO of the Center for AI Safety (CAIS) and founded the Center for AI Safety Action Fund.” Hadfield is “the Bloomberg Distinguished Professor of AI Alignment and Governance” at Johns Hopkins (AVERI, Team). See Bri Treece.
• The Forum seat. “AVERI is a founding member of the AI Evaluator Forum and some AVERI staff helped write their first standard, AEF-1” (Miles Brundage, The Launch of AVERI). The Forum’s members page lists AVERI first among eight members and names Patricia Paskov as Senior Advisor beside Chair Conrad Stosz (AI Evaluator Forum, Members). See AI Evaluator Forum.

Funding & Backers

• What AVERI says. “No donor makes up a majority of our funding. Our funders include the AI Underwriting Company, Coefficient Giving, Craig Falls, Geoff Ralston, Good Forever Foundation, Fathom, Halcyon Futures, Sympatico Ventures, and several non-executive employees and alumni of frontier AI companies.” On independence: “we do not accept cash donations or compensation from frontier AI companies or their executives” (AVERI, About). The launch post gives the same list and, in order, names Halcyon Futures and Fathom first (Miles Brundage, The Launch of AVERI). No amount, date or per-donor share appears on either page.
• Credits from the labs. “We’ve received API credits as donations or compensation from Amazon, Anthropic, Google DeepMind, Microsoft, OpenAI, and Thinking Machines Lab” (AVERI, About). The launch post, earlier, said “We’ve also been offered API credits from Amazon, Anthropic, Google DeepMind, OpenAI, and Thinking Machines Lab” (Miles Brundage, The Launch of AVERI).
• Conflicts disclosed. AVERI’s about page: “We also require public disclosure of the most significant conflicts on our website,” and lists, for Brundage: “AVERI’s Executive Director, Miles Brundage, earned equity in OpenAI while employed at the company, though he has since sold all of his shares,” with “Miles is recused from directly auditing OpenAI until two years after his October 2024 departure.” It adds that several employees hold broad stock indices and “individual investments in Microsoft, Google, NVIDIA, and Broadcom” (AVERI, About).
• The filing. None. The exemption dates from July 2025 and ProPublica holds no Form 990 data for the organization, so no revenue, donor-concentration or compensation figure is available from the IRS record (ProPublica Nonprofit Explorer).
• What the funders’ own pages add. Fathom: “We do not take donations from corporations, including frontier labs and the FAANG companies” (Fathom, 31 August 2026). Halcyon Futures: “Halcyon is a grant-maker, incubator, and investment firm. We back founders in AI safety, cybersecurity and biosecurity,” and its home page’s “Our Portfolio” carries links to averi.org, aiuc.com and transluce.org (Halcyon Futures, About; Halcyon Futures). The AI Underwriting Company’s name appears on AVERI’s list; aiuc.com, which Halcyon’s portfolio links, calls itself “a frontier AI standards & insurance company based in San Francisco” and reports a Series A: “The capital will help extend our audits, standards and insurance from agents to frontier models to create a strong ecosystem for frontier oversight” (AIUC; AIUC, Series A). The same company’s co-founder Rajiv Dattani is a listed author of AVERI’s flagship paper, under the affiliation “AI Underwriting Company” (AVERI, Frontier AI Auditing). Coefficient Giving, formerly Open Philanthropy, is named by AVERI; its own grants feed lists no AVERI entry, so the grant rests on AVERI’s statement (Coefficient Giving, search). See The EA to AI-Safety Funding Pipeline.

• A shared donor. Craig Falls, on AVERI’s funder list, is also a named supporter of Sentinel Bio, the biosecurity funder whose Treasurer, Max Henderson, sits on AVERI’s board (Sentinel Bio, Organization; ProPublica, Sentinel Bio FY2024 Form 990).

Institutional Voice & Intent

AVERI speaks in the register of a standard-setter that is explicit about its limits. Its founder pitches consensus: “it doesn’t make sense for companies to check their own homework on safety and security,” and “whether AI is a normal technology or an unusually dangerous one” the case holds (Miles Brundage, The Launch of AVERI). The flagship paper’s independence principle reads: “Auditors must be genuinely independent third parties, free from commercial or political influence”; “Safeguarding independence requires mandatory disclosure of financial relationships, standardized terms of engagement that prevent companies from shopping for favorable auditors, and cooling-off periods when moving, in both directions, between industry and audit roles”; “Alternative payment models that reduce auditor dependence on auditees should also be urgently explored” (AVERI, Frontier AI Auditing). The paper carries its own caveat: “nor does authorship imply an endorsement on the part of any individual’s organization.”

Stated intent: Make third-party audits of leading AI developers, with “deep, secure access to non-public information,” effective and universal, and be the standard-setter for what counts as a good audit and a qualified auditor (AVERI, About).

Observed intent: The same on the record. It coauthored the Forum’s evaluator standard, endorsed one state bill, supplied “expert input on other state and federal efforts,” published a flagship paper, a legislative landscape survey, a memorization-access paper and a pilot report, and ran a pilot with Google DeepMind (AVERI, About; AVERI, pilot report).

Gap: AVERI discloses its funders’ names, its lab credits, its Executive Director’s former equity and its recusal, and its paper states the independence rules it would apply to auditors. What its pages cannot show is the funders’ shares, since no donor amounts are published and no return is on file, or how the credits and the Google DeepMind pilot relate to the independence rules the paper sets. Its position is also its income stream’s precondition: a mission that is audits becoming universal. The record raises that and cannot settle whether it shapes what AVERI recommends; for the structural reading it does not need to.

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

1. Position: legally required, independent audits of frontier developers. “AVERI recently endorsed legislation for the first time, specifically HB 4705/SB 3261 in Illinois,” and “we believe that it would be reasonable to require 5-10 frontier AI companies to be audited starting in early 2027” (AVERI, Illinois post).
   • What adoption would pay it. A statutory buyer for the activity its mission names. The Illinois record describes the act that became Public Act 104-0538 as one that “mandates annual independent third-party audits” (ILGA, SB 315 status). AVERI states “We’re not trying to be the auditors exactly” yet describes “pilot audits,” and its team page gives one staff member the job of “building demand for frontier AI auditing” (Miles Brundage, The Launch of AVERI; AVERI, About; AVERI, Team). No fee, grant or contract to AVERI for any audit is documented; audit demand is the market of its funder the AI Underwriting Company, which describes itself as “a frontier AI standards & insurance company” and says its capital will “extend our audits, standards and insurance from agents to frontier models” (AIUC; AIUC, Series A).
   • Its best case. “Whether AI is a normal technology or an unusually dangerous one, it doesn’t make sense for companies to check their own homework,” and: “Currently, there is no state or federal requirement to do this in the US” (Miles Brundage, The Launch of AVERI; AVERI, Illinois post).
   • Who else gains from the same rule. Anthropic: AVERI says “We have been glad to see Anthropic also endorse this bill.” Every evaluator that sells or publishes audits; the insurers that price on them.
2. Position: audit standards written by the auditing field, then recognized in law. “We lead the development of frontier AI auditing standards, or the criteria for what counts as good audits and qualified auditors,” beginning with AEF-1, “a standard defining minimum operating conditions for independent third-party evaluations” (AVERI, About). Its Illinois post suggests legislation “designate one or more sources of formal standards,” naming CAISI, the Frontier Model Forum and “the AI Evaluator Forum (a group of assessment organizations)” as possible sources.
   • What adoption would pay it. The standard-setter’s seat. Its Standards Director is the Forum’s Senior Advisor and AVERI staff “helped write” AEF-1; a statute that cites the Forum cites a document AVERI coauthored (AI Evaluator Forum, Members; Miles Brundage, The Launch of AVERI). No fee for a standard is documented.
   • Its best case. “We need a more ambitious and more detailed “gold standard” for frontier AI audits”; the paper’s first challenge is “Ensuring high quality standards for frontier AI auditing, so it does not devolve into a checkbox exercise” (Miles Brundage, The Launch of AVERI; AVERI, Frontier AI Auditing).
   • Who else gains from the same rule. The Forum’s other seven members, which wrote and use the same text; the Frontier Model Forum; regulators that want a standard they did not have to write.
3. Position: a quality-control body for auditors, legal safe harbors for them, and audit-aware insurance. The launch post: “quality control for audits via a PCAOB-for-AI that combines benefits of the private sector, such as better-than-government salaries to attract top AI talent, with those of the public sector, such as democratic accountability and the ability to impose fines or pull necessary accreditations,” and “legal safe harbors for auditors that find things that companies don’t want to hear about” and “insurance pricing that takes auditing into account” (Miles Brundage, The Launch of AVERI).
   • What adoption would pay it. An accreditation layer over the auditors. AVERI says it will “likely be doing more of ourselves (though just using “soft pressure” rather than the formal authorities),” so adoption of a formal body would convert its present role into a statutory one; no seat on such a body is documented. AIUC, its funder, states audits, standards and insurance as its lines of work (AIUC, Series A).
   • Its best case. “We need to increase demand for audits enough to grow the ecosystem and mitigate risks at a sufficient pace, but also need to make sure that auditors aren’t churning out zillions of low-grade audits to meet unrealistic deadlines and turn a profit.”
   • Who else gains from the same rule. Fathom, whose sponsored statute, SB 813, has the state agency designate independent verification organizations; California’s AB 1405, which sets up a state registry of AI auditors; insurers; and the auditors, who would receive safe harbors.

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

AVERI sits beside the evaluators and one step upstream of them, with the sponsor of California’s evaluator-designation law among its funders. METR, Transluce and the other Forum members test models; AVERI argues that testing should be mandatory, independent and done to its standard. Fathom sponsored the statute that lets a state designate who may do the testing and funds AVERI. The AI Evaluator Forum is the table at which AVERI and the evaluators wrote the disclosure standard the audit would require.

• The California market AVERI has not said it will enter. SB 813, as chaptered on 9 September 2026, defines an IVO 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”; it provides that the act does not “Require any person, partnership, or corporation that develops, deploys, or operates an AI system or model to engage an IVO”; and an IVO “may accept payment from a party being assessed at reasonable market rates” (leginfo, SB 813 as chaptered). The companion AB 1405 would “prohibit an unregistered person from offering, selling, or conducting a covered AI audit” from 2029 (leginfo, AB 1405 as chaptered). AVERI’s own legislative survey of 20 April 2026 lists SB 813 as “Create a voluntary Independent Verification Organization (IVO) framework,” with “Government-certified IVOs would develop standards,” status “Pending,” and describes the Virginia study of an IVO framework “which would give liability benefits in exchange for participation in a voluntary audit regime,” linking Fathom’s IVO site; the one bill the post says AVERI endorsed is the Illinois bill (AVERI, Illinois post). No page read states that AVERI has applied, intends to apply or has taken a position on SB 813 or AB 1405. The statute defines an AI auditor as “a person, partnership, academic institution, nonprofit, or corporation that conducts a covered audit on behalf of a third party,” a class a nonprofit auditor such as AVERI would fall within if it conducted such audits.
• What the enrolled Illinois text requires. Public Act 104-0538 says “This Act takes effect January 1, 2027” but starts the audit later: “Beginning on January 1, 2028 or 90 days after a developer first qualifies as a large frontier developer, whichever is later, a large frontier developer shall annually retain a third party to perform an independent audit of compliance” with its framework and transparency section, so AVERI’s “2028” and ILGA’s 1 January 2027 describe different provisions of the same statute. The Act designates no auditor class and keeps no registry; the third party “shall possess demonstrated competence to perform the audit” and the developer “shall not condition any payment or the amount of any payment on the results of the third party’s audit” (Public Act 104-0538, as enrolled). ILGA lists SB 315 as filed in January 2025 as a technical amendment to the Predatory Loan Prevention Act and rewritten by Senate floor amendment on 11 and 19 May 2026, and lists HB 4705 and SB 3261, the bills AVERI endorsed, as separate bills last acted on 27 March and 22 May 2026; ILGA does not call SB 315 their descendant, AVERI’s note does (ILGA, SB 315 status; ILGA, HB 4705; ILGA, SB 3261; AVERI, Illinois post).
• Fathom’s design in AVERI’s own paper. The flagship paper lists Bri Treece of Fathom among its authors; its recommendation of “mandatory disclosure of financial relationships” for auditors and the Fathom-sponsored statute’s requirement that an IVO report annual changes to “governance policies or sources of funding relevant to the IVOs conflicts of interest or independence” address the same subject (AVERI, Frontier AI Auditing; leginfo, SB 813 as chaptered). The paper’s disclaimer applies: authorship “does not entail endorsement” by any author’s organization.
• Illinois. AVERI says it “provided feedback on and endorsed the legislation that became SB 315 in Illinois – the first US-based frontier AI audit requirement, which takes effect in 2028,” and its post’s note says the act “(a descendant of the bill we endorsed below) was signed into law and starts requiring audits in 2028.” The Illinois General Assembly record shows SB 315 as Public Act 104-0538, “Governor Approved” 6 July 2026, “Effective Date January 1, 2027” (AVERI, About; AVERI, Illinois post; ILGA, SB 315 status).
• The pilot. “In July and August 2026, AVERI conducted the world’s first double-blind evaluation of a proprietary language model in collaboration with Google DeepMind, OpenMined, and MLCommons,” testing Gemini 2.5 Flash-Lite with never-before-used MLCommons prompts inside a secure enclave; “We provided Google DeepMind with a confidential report of the findings” (AVERI, pilot report). AVERI says other voluntary pilots are under way: “we are currently pursuing pilots related to verifying that companies are following their own safety and security frameworks” (AVERI, About).
• Which others on the map are closely aligned, tie by tie.

Actions & Leadership Choices

CONDUCT: WATCH — ADVOCATE OF MANDATORY AUDITS, WITH DISCLOSED FUNDERS AND BOARD SEATS SHARED WITH THE LICENSING AND EVALUATOR NETWORK. No wrongdoing is alleged. AVERI states its funders, its credits and its Executive Director’s former equity, and sets its independence rules in public. The structural point is the one the Forum’s profile makes about every evaluator: the standard asks for disclosure, not absence. What is particular to AVERI is that it argues for the audit requirement that one funder’s stated lines of work build on, in the same policy field as the designation statute another funder sponsored, with no filing yet to show how much of its money comes from whom.

• 2025. Co-founded; IRS exemption issued July 2025 (Miles Brundage, The Launch of AVERI; ProPublica Nonprofit Explorer).
• 15 January 2026. Launches with the flagship paper, “coauthored with dozens of leading experts,” and a Fortune exclusive (AVERI; AVERI, launch announcement).
• 14 April 2026. Brundage on the Scaling Laws podcast, “Why AI Needs Independent Auditors” (AVERI, updates).
• 20 April 2026. Legislative landscape post and AVERI’s first bill endorsement, Illinois HB 4705/SB 3261 (AVERI, Illinois post).
• 27 April 2026. Post on near-verbatim extraction and the access auditors need, summarizing a preprint by advisor A. Feder Cooper (AVERI, updates).
• 6 July 2026. Illinois Governor approves SB 315, which AVERI calls a descendant of the bill it endorsed (ILGA, SB 315 status).
• 17 August 2026. Brundage on the Odd Lots podcast (AVERI, Odd Lots).
• 27 August 2026. Pilot report: the double-blind evaluation with Google DeepMind, OpenMined and MLCommons (AVERI, pilot report).

Alliance Map

Where ties on different axes run at the same time, the overlap is the finding. Four names on AVERI’s funders list recur on the evaluator map: Fathom sponsored the California statute that designates evaluators; Halcyon Futures’ chief executive sits on the boards of two Forum members and its portfolio page links a third organization on the list; Coefficient Giving is credited at three other Forum members; Geoff Ralston is on METR’s supporters list. The advisor tie runs the same way: Fathom’s President and Fathom’s senior advisor advise AVERI, and the senior advisor sits on the Governor’s expert group under the order that speeds the designation statute AVERI’s own survey lists as pending. The Standards Director who is the Forum’s Senior Advisor, the Executive Director who advises Epoch AI and RAND, and a board seat held by a Foundation for American Innovation officer extend the same set across the Forum, two research organizations and a policy organization. The documents show disclosed seats, listings and funding, not coordination, and no designation has been made.

Reach Assessment

Institutional: Moderate and conditional. AVERI cannot require an audit, designate an auditor or certify a result. Its reach is a paper the field cites, a standard the Forum’s members attach to their reports, a bill it describes as the predecessor of an Illinois law and a pilot that put a lab’s proprietary model through an outside test, all voluntary on the lab’s side.

Memetic: Moderate. “Checking their own homework” and the comparison to the accounting profession’s oversight board (PCAOB-for-AI) are AVERI’s frames, and they travel in policy discussion alongside the IVO term of art, which AVERI’s own survey notes.

Civilizational: Moderate, with low durability. If audit requirements spread from Illinois, the people who write the audit standard and the quality-control rules sit upstream of whether any developer’s safety claims are believed, with a funder base that includes the designation statute’s sponsor and an unfiled return. AVERI’s pages say the right things about that: published funders, a conflicts policy, a recusal and the paper’s own independence rules. The reach is the finding; the malice is not asserted.

Score basis. Atk 5 is reach: AVERI’s about page says it coauthored AEF-1 and endorsed the Illinois bill, and SB 315, which AVERI calls its descendant, was approved on 6 July 2026 with audits required from 2028; the Forum says the EU AI Office has endorsed key provisions of AEF-1; and AVERI ran a pilot evaluation of Gemini 2.5 Flash-Lite with Google DeepMind (AVERI, About; ILGA, SB 315 status; AI Evaluator Forum, AEF-1; AVERI, pilot report). It is not higher because no statute or procurement rule makes any lab ask for its work, its page says “Frontier AI audits aren’t yet required,” and it has published pilot summaries and papers, not an audit of a lab. It sits level with the AI Evaluator Forum (5/6/4, hazard 50), whose standard it coauthored and whose reach rests on the same endorsement, and with Apart Research (5/5/5, hazard 50); one below Transluce (6/7/6, hazard 52), whose tools sit in a lab’s training review and whose findings a lab’s system card cites; and two below FAR.AI (7/6/6, hazard 59), whose red-teaming engagements test deployed models. Def 6 is funding insulation: no cash from frontier AI companies or their executives, no majority donor, a named funders list and a disclosed recusal (AVERI, About). The points off are that no amounts or shares are published and no return is on file (ProPublica Nonprofit Explorer), that donors include non-executive employees and alumni of frontier AI companies, that credits came from six AI companies including its pilot partner, that its named funders recur behind other organizations on the map, and that Fathom, a funder, sponsored the designation statute. Peers on def 6: the Forum, FAR.AI and Palisade Research (7/6/6); Transluce holds 7 because it publishes its revenue shares and has a filed return. Hp 4 is entrenchment: an IRS exemption from July 2025, seven named staff, a board of three and six advisors, a flagship paper, a coauthored standard and an endorsed bill that became law (AVERI, Team; AVERI, Frontier AI Auditing). The points off are that no statute or contract installs it, no return is filed, the record is about fifteen months, and its mission depends on audit requirements its own page says do not yet exist beyond Illinois. Peers on hp 4: the Forum (4); Apart Research holds 5 on 59 Sprints and a fellowship pipeline, and Transluce holds 6 on an installed open-source tool and a contractor seat with the EU AI Office. The values hold at 5/6/4, hazard 50.

Sources: AVERI — home; AVERI — About; AVERI — Team; AVERI — Updates; AVERI — Frontier AI Auditing; AVERI — Frontier AI Auditing-Related Legislation in the US; AVERI — Pilot Report: double-blind evaluation; AVERI — launch announcement; AVERI — Odd Lots Podcast; Miles Brundage — The Launch of AVERI; ProPublica — Ai Verification And Evaluation Research Institute; ProPublica — Halcyon Futures Inc; ProPublica — Fathom AI Inc; AI Evaluator Forum — Members; AI Evaluator Forum — AEF-1; Fathom — About; Fathom — Partnerships; Fathom — SB 813 passes; Fathom — Hadfield appointed; Halcyon Futures; Halcyon Futures — About; AIUC; AIUC — Series A; Transluce — About; METR — About; Princeton HAL; SecureBio — GPT-6 Astra assessment; Coefficient Giving — grants search, RAND; Coefficient Giving — search, AVERI; Governor — experts announced, 23 September 2026; leginfo — SB 813 text; leginfo — AB 1405 text; ILGA — SB 315 bill status.