ARTIFICIAL INTELLIGENCE UNDERWRITING COMPANY (AIUC)

OLYMPUS opened an institutional file on the Artificial Intelligence Underwriting Company because it is where the audit gets a price. The evaluators on this site’s map (METR, Transluce, AVERI, Vals AI) produce findings; a finding becomes mandatory without a statute only when someone who pays for the damage reads it. AIUC is a company that says in its own words that this is its trade: it writes a standard, conducts the testing behind a certificate, and reports that insurers offer coverage to the companies that earn the certificate. It is a for-profit San Francisco company with $55 million from named venture investors, and it is the one body on the evaluator map built to turn an evaluation into an insurance term. The finding is the position: a standard-setter, a certifier and an insurance channel in one firm, whose co-founder sits on the board of the grader that frontier labs hand their models to, and whose funders overlap the incubator behind two other evaluators.

Institutional Archetype

THE UNDERWRITERS LABORATORY FOR AI. AIUC adopts the oldest argument in product safety, and states it on its home page: “Electricity burned down many houses until insurers funded Underwriters Laboratories (UL) to certify lightbulbs.” Its claim is that “The combination of standards, audits, and insurance creates an incentive flywheel that has taught the world to trust electricity, cars and nuclear power” (AIUC). It does not build models and does not write refusals. It writes the test, runs the test, and reports what the test is worth to an underwriter. The instrument is the certificate; the product is the term on which someone will insure the thing certified.

Mandate & Origin

AIUC describes itself as “a frontier AI standards & insurance company based in San Francisco,” and its team page gives the mission and the premise: “Our mission is to Underwrite Superintelligence. We believe progress requires security, incentives must be shaped, AI risk is a new version of an old problem, and markets can help govern AI” (AIUC, team). Its earliest dated post is a paper on AI-proofing boards and executive teams, co-written by Rajiv Dattani with an MIT Sloan researcher, dated 21 July 2025 (AIUC, AI-Proofing The Board and C-suite). Reinsurance News reported the launch on 23 July 2025 with a $15 million seed round, “the largest seed round in insurance history” in the company’s words as the outlet reports them (Reinsurance News).

• Name. The company’s own releases read “The Artificial Intelligence Underwriting Company” (ElevenLabs) and “the Artificial Intelligence Underwriting Company (AIUC)” (KPMG); aiuc.com abbreviates it AIUC; AVERI’s funder list and AVERI’s paper write “AI Underwriting Company” (below).
• People. The team page lists two co-founders, Rune Kvist, “First product hire at Anthropic,” and Rajiv Dattani, “Board member @ METR. Former McKinsey Insurance partner” (AIUC, team). ElevenLabs’ release gives Kvist’s title: “Co-founder & CEO of The Artificial Intelligence Underwriting Company” (AIUC, ElevenLabs). The team page also says the staff are “coming from METR, Anthropic, OpenAI, Allianz and McKinsey Insurance.” Reinsurance News names a third co-founder and CTO, Brandon Wang, and describes the founders’ background in the Lloyd’s market and at Anthropic and the Center for AI Safety, as reported; AIUC’s own team page lists no third co-founder.

The standard. AIUC-1 is described on its release as “the security, safety and reliability standard for agents,” “built with 250+ security and risk leaders and frontier AI builders like Cursor, ElevenLabs, and Harvey” (AIUC); the standard’s own site now counts “300+ Fortune 500 security leaders” and says it is “Maintained by Artificial Intelligence Underwriting Company,” updated quarterly (AIUC-1). The Series A post gives the mechanics: “It tests how well agents hold up against risks such as jailbreaks, hallucinations and data leaks, using 5,000 risk-and-attack combinations unique to each type of business” (AIUC, Series A). Six domains, A to F: data and privacy, security, safety, reliability, accountability, society; the last carries “Cyber Misuse” and “Catastrophic Risk” (AIUC-1). The standard is cross-walked to ISO 42001, the NIST AI RMF, the EU AI Act, MITRE ATLAS and the OWASP Top Ten; its site says “AIUC-1 turns high-level frameworks and legislation into auditable controls.”

Who does the auditing. On 3 February 2026 Schellman became “the first authorized auditor of AIUC-1.” The division of labour is in the release AIUC hosts: “Under this partnership, the Artificial Intelligence Underwriting Company conducts technical evaluations and issues certification, while Schellman provides independent audit evidence collection, detailed reporting, and certification guidance” (AIUC, Schellman). KPMG LLP, certified on 27 August 2026 for its aIQ Capture platform, is described as “the first of the Big Four to achieve AIUC-1 certification” and joined the AIUC-1 Consortium (AIUC, KPMG). The company that writes the standard also conducts the evaluations and issues the certificate; an outside accounting firm collects audit evidence.

Funding & Backers

AIUC publishes its investors and its totals but not its revenue.

• Investors, in its own words. “We’ve raised $55M from Nat Friedman, Ribbit Capital, First Harmonic and Terrain” (AIUC, team). The Series A post, 15 September 2026: “We’ve raised $40 million Series A led by Ribbit Capital, with participation from First Harmonic and Terrain. Along with our $15 million seed round from NFDG, this brings our total funding to $55 million” (AIUC, Series A). Trade press adds seed participants Emergence and angels including “Ben Mann, co-founder of Anthropic, and former chief information security officers at both Google Cloud and MongoDB,” as reported (Reinsurance News); SiliconANGLE reports the same round (SiliconANGLE, 15 September 2026).
• Halcyon Futures’ page. Halcyon Futures carries a testimonial from Dattani under the labels “Career Transition Grant,” “Halcyon Retreat,” “Invested in Every Round” and “$16M raised from top VCs”: “Halcyon has been instrumental for my journey to founding AIUC – If I were starting something new, Mike would be my first call” (Halcyon Futures). The labels and the $16 million figure are Halcyon’s. AIUC’s own pages name four investors and give a $15 million seed; they do not name Halcyon, and no page gives Halcyon’s stake or amount.
• Customers. The Series A post lists Cursor, ElevenLabs, Harvey, KPMG, Lovable, UiPath and Fin as certified; AIUC-1’s site adds Sierra, and MSCI as an adopter. No fee, price or revenue figure appears on any page read. Certification is bought by the company certified.
• AVERI funding. AVERI’s about page: “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” (AVERI, About). AVERI’s flagship paper lists Dattani, with affiliation “AI Underwriting Company” (AVERI, Frontier AI Auditing). Neither AVERI nor AIUC states that the funder and aiuc.com are the same company, and no amount appears; the name and Dattani’s listed affiliation are the match.

Concentration. Four named investors and no frontier lab on any AIUC page. The conflict arrives from the other side of the ledger: the customer pays for the grade. Nothing here says a customer has influenced a result; the pages show who buys the product and give no price.

Institutional Voice & Intent

AIUC’s register is the enterprise-security brochure fused with the policy essay: confident, historical, thick with named customers. The Series A post opens “Pandora’s box is open” and calls for “auditors with frontier AI expertise, incentives that prevent a race to the bottom, and standards for what good audits look like across the industry” (AIUC, Series A).

Stated intent: “we build the audits, standards, and insurance infrastructure that allow enterprises to deploy AI with confidence” (AIUC, Series A); extend “from agents to frontier models to create a strong ecosystem for frontier oversight.”

Observed intent: Make AIUC-1 the certificate enterprise buyers ask for: “Ask your AI vendors if they are AIUC-1 certified. The 50+ page independent audit report covers how guardrails are implemented and red-teaming results” (AIUC-1). Tie that certificate to insurance terms. Move up from agents to the frontier labs themselves.

Gap: Small in word, structural in effect. AIUC says the testing is the product and the insurance follows it. The reported form of the link: “The cost and availability of insurance are tied directly to audit results: safer systems are offered better terms,” in the company’s launch statement as Reinsurance News reproduces it (Reinsurance News). ElevenLabs’ release says its test scenarios generate “the empirical risk profile insurers need to underwrite AI,” and that testing “has enabled leading insurers to offer an AI-specific insurance policy” (AIUC, ElevenLabs). No page names the insurer, the premium or the share of risk the insurer carries. The record does not settle whether AIUC’s certificate moves any price; the pages state that it does, name no carrier, and give no number.

Interest: Three positions, each in the company’s own words.

• Insurance priced on audit results. Position: “The combination of standards, audits, and insurance creates an incentive flywheel” (AIUC); co-founder Dattani, on the ElevenLabs policy: “Certifying against AIUC-1 gave insurers the confidence they needed to offer @elevenlabs first-of-its-kind voice agent insurance” (Dattani, X). What adoption would pay it: every enterprise buyer that asks vendors for the certificate is a customer for the audit AIUC conducts and issues; AIUC-1’s site says so (“Ask your AI vendors if they are AIUC-1 certified”) (AIUC-1). Fees are not documented. Best case, AIUC’s: “Electricity burned down many houses until insurers funded Underwriters Laboratories (UL) to certify lightbulbs” (AIUC), and a certificate that costs something to fake is worth more than a vendor’s self-attestation. Who else gains: the auditors who qualify (Schellman, KPMG), the vendors that want a trustmark, and the security teams that obtain a purchasing tool.
• A frontier-lab mutual insurer. Position, from the AIUC post by Kvist and the Lawfare article by Cristian Trout, Kvist and Dattani: “Build a frontier AI mutual insurance company to hold the risk,” “Owned by the frontier AI companies it covers” (Lawfare, 22 September 2026). Audits there are “selected, contracted, and paid for by AIMIL” and “conducted by qualified third parties such as Accenture or METR”; “If elevated risks are discovered, premiums increase and remediations are recommended.” What adoption would pay it: a standing buyer for audits and evaluations, “a market for third-party assurance,” in AIUC’s own words, and a body staffed with “underwriters and actuaries specialized in AI risk.” AIUC’s lines of work are audits, standards and insurance; neither page says AIUC would run, staff or supply the mutual. Best case, theirs: “Mutuals have skin in the game,” “Mutuals have teeth” (AIUC, Frontier AI Mutual Insurer), and “It chooses the evaluators, so labs can’t shop referees.” Who else gains: the evaluators named as example contractors (METR, whose board Dattani sits on, and Accenture), and the labs, whose pooled liability becomes a priced and bounded line.
• Private verification first, a mandate later. Position: “We claim only that the private sector can and should lead, developing and testing which mitigations work in practice before codifying them in law,” and “If the regime proves itself, it could be strengthened and spread by mandating insurance, as we do for countless other activities” (Lawfare). What adoption would pay it: a mandated insurance requirement is the rule that turns a voluntary certificate into a condition of operating; whoever supplies the evaluations the insurer prices on is upstream of that condition. Best case, theirs: private verifiers and labs “can experiment now” (Dattani, X), and nuclear insurance, in Dattani’s words, “works this way: insurers employ nuclear engineers who inspect facilities” (Dattani, X). Convergence: AVERI argues for mandatory third-party audits and lists AIUC among its funders; Fathom sponsors the California law that designates independent verification organizations; the three are mapped on The Evaluator Alliances.

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

The chain, with each link on a page one of the parties published. A METR board member co-founded an insurer that prices AI risk, backed by the incubator whose chief executive sits on the boards of AVERI and Transluce.

1. METR. METR’s about page lists Rajiv Dattani as “Advisor and Board Member”; his METR team page says he is “building AI auditing and insurance mechanisms at the Artificial Intelligence Underwriting Company” and that his previous roles include “COO at METR, Founding Advisor at Transluce, and Partner at McKinsey” (METR, About; METR, Rajiv Dattani). AIUC’s team page gives the same seat: “Board member @ METR.”
2. The insurer’s essays name METR. The Lawfare article, co-signed by Dattani (Lawfare), names METR twice as a third-party contractor an insurer would engage (“a third party such as METR to help OpenAI contain the incident”; “qualified third parties such as Accenture or METR”). METR’s own conflict of interest policy states, for payments, “To date, we have not received payment for work on company-identifying risk assessments”; its report on the OpenAI and Hugging Face incident says “Per our standard policy, we did not take payment from OpenAI for this independent assessment” (METR, conflict of interest policy v1.1; METR, incident report). No page read states that METR has been engaged or paid by AIUC or any insurer.
3. The same incident, two accounts. AIUC’s mutual post: “Today, OpenAI picks the auditor, sets the investigation’s scope, and keeps most of its lessons private.” METR’s report of that investigation says “OpenAI defined the investigation period as June 26th through July 13th” and “On July 29th we reached an agreement with OpenAI to conduct an independent investigation of the Hugging Face attack scoped to the following seven questions,” one of which “was added at the request of OpenAI.” The two texts agree on the arrangement. One is written by the insurer’s co-founders and the other by the grader whose board one of them sits on.
4. Halcyon. Halcyon Futures’ page carries Dattani’s testimonial credit for AIUC’s founding and the label “Invested in Every Round.” Its founder and chief executive, Mike McCormick, is a board member of AVERI (“Mike is a board member at AVERI and the founder and CEO of Halcyon Futures”) and is listed as a director of Transluce (“Mike McCormick CEO, Halcyon Futures”) (AVERI, team; Transluce, About).
5. AVERI. AVERI lists “the AI Underwriting Company” first among its funders; the flagship paper lists Dattani and Trout under that affiliation, beside METR’s Chris Painter and Transluce’s Conrad Stosz (AVERI, About; AVERI, Frontier AI Auditing). Dattani on AVERI’s launch day: “We need 3rd party audits & verification to unlock the insurance market and create confidence in AI adoption. Excited for AVERI’s work!” (Dattani, X).
6. Transluce. Dattani’s METR page lists him as a former Founding Advisor at Transluce; on 24 September 2026 he posted: “3 new hacks by OpenAI agents discovered in a project led by @TransluceAI where AIUC contributed” (Dattani, X).
7. The Forum. AIUC is not one of the AI Evaluator Forum’s eight members. Dattani and Tom Fehring, listed as “Risk, Artificial Intelligence Underwriting Company (AIUC),” are among the 200-plus signatories of the Forum’s 18 September 2026 letter, whose minimum conditions include that embedded evaluators “should not have other significant commercial business with” frontier AI companies “and should not accept any form of payment or other reward contingent on the evaluator’s findings” (AI Evaluator Forum, letter). The chief executives of Armilla AI and Vals AI sit among the signatories as well.

The other places in the apparatus.

• Anthropic. Kvist is described on AIUC’s team page as “First product hire at Anthropic”; the team page lists Anthropic and OpenAI among the staff’s former employers; the AIUC-1 consortium list prints “Jason Clinton Deputy CISO at Anthropic” among its members (AIUC, team; AIUC-1, consortium). See Anthropic.
• The insurance and audit trade. The consortium list, as printed, includes executives from Marsh, Aon, Liberty Mutual, CFC, New York Life, Deloitte and KPMG alongside the 296 members. They are consortium members, not named carriers of any AIUC policy.
• Standards bodies. AIUC-1 is “designated CSA STAR Level 2 alongside ISO/IEC 42001” (Cloud Security Alliance, 1 July 2026); IBM Research integrated it into its AI Risk Atlas (12 March 2026); MITRE became “a technical contributor” (9 December 2025) (AIUC-1, CSA; AIUC-1, IBM; AIUC, MITRE ATLAS).
• The peers on the map. AVERI advocates the audits AIUC would price; Artificial Analysis and Vals AI sell benchmarks to the same customer base; Halcyon Futures seeds the layer; Fathom sponsors the state designation law. AIUC alone sits at the insurance end of the chain.

Actions & Leadership Choices

The PR says “confidence infrastructure.” The deeds are a sequence of certifications, an accreditation, a standard re-issued each quarter and, since September 2026, a pivot to the frontier.

• A certificate that insurers are said to read (11 February 2026). ElevenLabs became “the first company to go live with the AIUC-1-backed insurance policy covering AI voice agents”; its agents underwent “more than 5,000 adversarial simulations” and “leading insurers” offer coverage (AIUC, ElevenLabs). The insurers are unnamed on AIUC’s page and on ElevenLabs’ own announcement.
• Quarterly re-issue. The standard’s site lists releases for Q1 through Q4 2026, the last, dated 15 October 2026, titled “Open-weight model controls and stronger third-party testing” (AIUC-1). Dattani’s post of 16 October 2026: “We worked with @baseten and 100+ Fortune 1000 security leaders to update the AIUC-1 standard” (Dattani, X).
• The pivot to frontier models (15 September 2026). “The capital will help extend our audits, standards and insurance from agents to frontier models” (AIUC, Series A). On 26 June 2026 Dattani posted: “We’re building private market standards & audits for frontier labs — combing technical expertise with trusted audit practices. Who wants to build it with us?” (Dattani, X).
• The mutual proposal (22 and 23 September 2026). Published on Lawfare and on AIUC’s site within a day, with the claim that “the first policy could be issued within 3 months” (AIUC, Frontier AI Mutual Insurer). “A mutual would still have limitations,” its post concedes: “A mutual’s capital could cover only a fraction of a major catastrophe.”
• A report on the insurance stack (July 2026). GovAI lists “Underwriting the Agent Economy: The Blueprint for an AI Insurance Stack,” dated 22 July 2026, with an author list that begins with Cristian Trout and includes Markus Anderljung; its abstract argues “that affirmative AI coverage with limits in the billions is achievable by 2030, but only with industry-wide coordination” (GovAI). Dattani’s post of 14 July calls it “Our new report” and thanks co-authors including Miles Brundage of AVERI (Dattani, X; Dattani, X).

CONDUCT: WATCH — STANDARD-SETTER, CERTIFIER AND INSURANCE CHANNEL IN ONE COMPANY, WITH A CO-FOUNDER ON THE GRADER’S BOARD AND ITS NAME ON AVERI’S FUNDER LIST. The record shows a company that conducts what it describes as the testing, issues what it describes as the certificate, tells buyers the certificate unlocks insurance, and now proposes to price the frontier labs. It also shows seats and funding lines the parties disclose. It does not show a certificate that was given or withheld for any reason other than the stated test, a price paid by any insurer on any certificate, or any influence over METR’s work.

Alliance Map

These run at the same time. One company writes the standard, conducts the testing behind the certificate, is read by a consortium of about three hundred security leaders that includes insurers and a Big Four auditor, tells buyers that insurers offer coverage to the certified, proposes a lab-owned insurer that would hire the evaluators it names, is named as a funder by the advocate of mandatory audits, and has a co-founder on the board of the grader the same labs hand their models to, a founding-advisor listing at a second evaluator, and a career-transition grant and investment credit on the incubator page whose chief executive sits on the boards of two more. Each tie is disclosed by AIUC, by METR, by AVERI or by Halcyon. The overlap is the finding; the documents show disclosed seats and funding lines, not coordination.

Reach Assessment

Institutional: Moderate and rising. AIUC-1 sits in enterprise security teams’ control libraries and in the Cloud Security Alliance registry, and a Big Four firm, a coding-agent company, a voice-agent company and a legal-agent company hold the certificate. No regulator, statute or lab pre-release process compels it. The frontier step is a stated plan, not a record.

Memetic: Moderate. The UL analogy is the page’s argument and it travels: insurers funding the test that becomes a trustmark is an easy story to retell, and AIUC’s Lawfare article and X threads retell it with nuclear, boiler and automobile precedents. When the conversation about AI oversight is carried out in the vocabulary of premiums, audits and “skin in the game,” it is carried out in vocabulary AIUC uses.

Civilizational: Low to moderate, conditional. The mutual proposal would, if adopted, put a lab-owned insurer with audit and suspension powers on the pre-deployment path; it has not been formed, and its authors say it “could cover only a fraction of a major catastrophe.” Insurance is how an audit becomes a condition of operating without a statute; AIUC is the firm on the map that sells that conversion.

Score basis. Atk 6 is reach. AIUC’s Series A page lists Cursor, ElevenLabs, Harvey, KPMG, Lovable, UiPath and Fin as certified against AIUC-1; ElevenLabs’ release records the first AIUC-1-backed insurance policy going live on 11 February 2026; and the Cloud Security Alliance designates AIUC-1 for its STAR registry beside ISO/IEC 42001 (AIUC, Series A; AIUC, ElevenLabs; AIUC-1, CSA). The points under a higher number are that nothing obliges a buyer, a lab or a regulator to use it and no carrier is named. Def 4 is funding insurance: four named investors and a $55 million total, no frontier lab among them; customers buy the certification AIUC conducts and issues and no fee is published; AIUC also names itself among AVERI’s funders (AIUC, team; AVERI, About). Hp 4 is entrenchment: the standard is re-issued each quarter, CSA STAR and the IBM Risk Atlas carry it and Schellman audits it; the company’s earliest dated post is July 2025, no statute or multi-year contract is documented, and the insurer behind the policy is unnamed (AIUC-1; AIUC, Schellman). Peers on the same values: Vals AI at 6/4/5, hazard 50, holds the same reach and funding insulation as a for-profit evaluator whose customers pay to be tested, one point more entrenchment from a $40 million Series A and government work; AVERI at 5/6/4, hazard 50, holds the same entrenchment, a point less reach and two points more funding insulation because it takes no money from the graded; Artificial Analysis at 8/4/6, hazard 60, holds the same insulation and far more reach because the labs quote its index at launch; Halcyon Futures at 5/4/3, hazard 40, the seed funder AIUC’s co-founder credits, holds the same insulation with less reach and entrenchment. The values hold.

Sources: AIUC; AIUC, team; AIUC, Series A; AIUC, Frontier AI Mutual Insurer; AIUC, ElevenLabs; AIUC, Schellman; AIUC, KPMG; AIUC, MITRE ATLAS; AIUC, AI-Proofing The Board and C-suite; AIUC-1; AIUC-1, consortium; AIUC-1, CSA STAR; AIUC-1, IBM Research; AIUC-1, CRI; Lawfare, Bootstrapping Frontier AI Governance by Mutualizing Risk; GovAI, Underwriting the Agent Economy; AVERI, About; AVERI, Team; AVERI, Frontier AI Auditing; Halcyon Futures; Transluce, About; METR, About; METR, Rajiv Dattani; METR, conflict of interest policy v1.1; METR, OpenAI / Hugging Face incident report; AI Evaluator Forum, Minimum Conditions for Embedding Evaluators; Dattani on X, @RajivDattani; Reinsurance News, 23 July 2025; SiliconANGLE, 15 September 2026.