CONRAD STOSZ

Behavioral Archetype

THE POLICY DRAFTSMAN WHO CROSSED THE TABLE — Subject spent his career before Transluce on the government side of the question of who tests frontier models, in the Pentagon, on a House intelligence committee, at the White House budget office and at the national standards institute, and now writes the rules that outside evaluators say they will hold themselves to. Transluce’s announcement says he “previously led the U.S. Center for Standards and Innovation.” The same document names his present job: “lead efforts to define standards and best practices for AI evaluations.” He is also the Chair of the AI Evaluator Forum, the network of evaluators that wrote the standard. The seat moved from the grader of the labs’ voluntary submissions to the graders’ own trade standard. This profile scores the seat, not the man.

Essence Indicators

• Head of Governance, Transluce. “Transluce has named Conrad Stosz as its first Head of Governance.” The announcement is dated 22 October 2025; his About-page entry reads “Head of Governance” beside Chief of Staff Tiffany Tzeng (Transluce, Head of Governance; Transluce, About). The employer is Transluce, led by Jacob Steinhardt and Sarah Schwettmann.

• What the job is, in the employer’s words. “At Transluce, Conrad will lead efforts to define standards and best practices for AI evaluations and inform the continued development of AI policy” (Transluce, Head of Governance).
• Chair, AI Evaluator Forum. The Forum’s members page lists “Conrad Stosz” under Leadership as “Chair,” with Patricia Paskov as Senior Advisor. The Forum describes itself as “a network of organizations conducting independent, third-party evaluations of AI systems”; the members page lists AVERI, the Collective Intelligence Project, Meridian Labs, METR, the Princeton Holistic Agent Leaderboard, RAND, SecureBio and Transluce, and says membership “is limited to organizations that conduct and publish rigorous independent technical evaluations of general-purpose AI systems in the public interest.” Its home page adds: “The Forum is not a legal entity in its own right and relies on voluntary participation from its members” (AI Evaluator Forum, members; AI Evaluator Forum). Stanford’s Center for AI Safety lists him the same way: “Conrad serves as Head of Governance for Transluce, a nonprofit AI research lab, and Chair of the AI Evaluator Forum, a network of independent AI evaluation organizations” (Stanford Center for AI Safety, SAFE Annual Meeting 2026). The Associated Press says he “chairs the AI Evaluator Forum, which is drafting best practices for the field” (AP via PBS NewsHour, 28 September 2026).

• The public launch. A Transluce-hosted event listing for “the public launch of the AI Evaluator Forum, a collaborative network of leading independent AI evaluation organizations working in the public interest,” lists Subject as a host and a speaker: “Conrad Stosz, former Acting Director of the U.S. Center for AI Standards and Innovation.” The listing says the event would “release a new standard for independent evaluations” (Luma, AI Evaluator Forum). The Forum’s AEF-1 page dates the standard “Version 1, updated December 4, 2025” (AI Evaluator Forum, AEF-1).

• Before Transluce: the standards institute. Transluce says he “previously led the U.S. Center for Standards and Innovation, as well as leading AI policy development in a range of positions in the White House, Congress, and Department of Defense, building on his prior experience as a machine learning engineer” (Transluce, Head of Governance). The name in that sentence omits “AI”; Transluce’s own event listing and the AP use the Center’s name: “Center for AI Standards and Innovation,” or CAISI (Luma; AP via PBS). The federal body is the one profiled at US CAISI. His earlier post there was Head of Policy: his own LinkedIn announcement, as MeriTalk quotes it, read “Thrilled to share that I have started a new position as Head of Policy for the U.S. AI Safety Institute in the Department of Commerce, wrapping two wonderful years leading on federal AI policy from the White House Office of Management and Budget” (MeriTalk, as reported). MeriTalk dates the move to August 2024 and says he joined the institute on a temporary assignment in April.

• Before the institute: the White House budget office. His own bio supplied for a GovExec event reads: “Conrad Stosz is the Director of Artificial Intelligence in the Office of the Federal Chief Information Officer (OFCIO) at the White House,” focused on “developing and issuing policy to advance federal agencies’ use of AI to improve customer experience and support agency missions, while protecting the public from AI risks” (GovExec event bio, PDF). MeriTalk reports that he “began his work at OMB in July 2022 as a policy advisor,” and quotes him on the Biden-era order: “I am also tremendously grateful to have had this opportunity over the last two years to help shape the U.S. response to recent advances in AI, including the chance to help shape Executive Order 14110 and its implementation” (MeriTalk, as reported). Executive Order 14110 was revoked on 20 January 2025, as recorded at US CAISI.

• Before that: Defense, Congress, the intelligence community. The same bio: “Previously, he was an AI Policy Advisor for the Office of the Secretary of Defense and oversaw a multi-billion dollar intelligence program for the House Permanent Select Committee on Intelligence,” and “He also served as a policy fellow for the Defense Innovation Board and built data science tools for the Open Source Enterprise within the Central Intelligence Agency’s Directorate of Digital Innovation.” It adds an M.A. in Security Studies from Georgetown University and a B.A. in International Relations from Boston University (GovExec event bio, PDF). MeriTalk puts the Defense and Capitol Hill years at “just under four years” and says he spent “more than six years in industry” before federal service (MeriTalk, as reported).

• The conference talk. Stanford’s October 5, 2026 SAFE Annual Meeting programs “Conrad Stosz · Transluce” for an invited talk titled “Improving the science and practice of evaluating AI companions” (Stanford Center for AI Safety).

Social Persona / Impression Management

Immediate impression: The procedural insider. Every bio line is a job title in a policy office, and his quoted remarks are about process: who gets access, who is independent, what a word means.

Energy: Quiet public footprint. No personal account was found on any platform; the Transluce site’s footer links only the organization’s accounts. His public voice is the Transluce announcement, the Forum’s pages, a conference listing and two press quotes.

Impression management strategy: Convert the standard into the argument. When the AP asked about the labs’ embedded-evaluator pledges he did not take a side on the pledges; he asked what the term covers: “Lots of evaluators are interested in embedding with labs and getting greater access, but it’s a little ambiguous what embedded evaluators means,” and “Will evaluators be able to thoroughly investigate, assuming that access is granted in a way that does not undermine their independence and credibility?” (AP via PBS). On the Forum’s letter he framed the goal as common ground: “We’re just trying to really demonstrate a shared common ground on basic principles and ensure that independent oversight can be a meaningful tool for managing AI risk broadly,” as IBTimes reports him telling CNBC (IBTimes, as reported). The definition is the argument: whoever sets what “embedded evaluator” means sets what the pledge delivers.

Forensic Archetype Comparison

Psychometric Assessment

Trait scores are observations of the documented public role, not claims about private character.

Big Five (OCEAN):

Dark Triad (held low and evidence-bound; the score measures structural position, not character):

MBTI: INTJ (“The Architect”) — a standards drafter by trade, who answers press with definitions and process questions, and moves between institutions to keep working on the same design problem. Rationale is the public record, not a typing session.

Threat Assessment

Score move: none (new profile). Calibrated against Sarah Schwettmann (5/6/5, hazard 43, co-founder and Chief Scientist of the same employer) and Jacob Steinhardt (6/7/6, hazard 50, co-founder and CEO), with Chris Painter (6/8/7, hazard 62, President of METR, a Forum member, with a Senate witness seat) and Elizabeth Kelly (7/8/7, hazard 72, the founding director of the U.S. institute he later worked in) above. He holds the same leverage as Schwettmann through a different instrument and less insulation, because he is staff and not a co-founder; 5/5/5 and 40 place him one point under her on def and three under on hazard.

Score basis. Atk 5 is leverage: he chairs the Forum behind AEF-1, which the EU AI Office has endorsed “key provisions of” for the independence provisions of the General-Purpose AI Code of Practice, and he leads governance at the evaluator that is a Forum member (AI Evaluator Forum, AEF-1; Transluce). It is not a 6 because the Forum “is not a legal entity in its own right,” and AEF-1 itself says its success “requires adoption by a range of stakeholders, including frontier developers.” Def 5 is insulation: Transluce “does not accept any funding that would grant a funder editorial control over its outputs” and “has carried out no paid evaluations,” and no board or court has overruled him (Transluce, policy). It is not higher because the policy discloses 6 percent of revenue from the personal holdings of OpenAI employees and 32 percent from those of Anthropic employees, and because he holds no founder control or tenure. Hp 5 is durability: AEF-1 is a published standard with an endorsement by an EU body and use in published reports (AI Evaluator Forum, AEF-1). It is not higher because nothing in statute installs it and the Forum has no legal personality.

Alliance Map

The overlap is the finding. The Agency career, the Corporate seat at an evaluator whose revenue includes lab employees’ personal holdings, and the Movement chair of the network that writes the evaluators’ standard all run at the same time, and they converge on one product: a published standard for independent evaluation, written by the people who do the evaluating, referred to by a regulator and addressed to the labs. All of it is in the open. Transluce’s own policy says it is written to be consistent with AEF-1, and its mental health evaluation is listed on the AEF-1 page as an instance of the standard in use.

Alignment Analysis

Stated alignment: “At Transluce, Conrad will lead efforts to define standards and best practices for AI evaluations and inform the continued development of AI policy” (Transluce). The Forum he chairs states its purpose as advancing “the rigor, credibility, and impact of independent AI evaluations that serve the public interest” (AI Evaluator Forum, mission).

Observed alignment: He moved from federal AI policy to the nonprofit that tests the labs’ models and chairs the network whose standard sets the terms for that testing. The Forum’s letter asks that “all frontier AI companies should embed evaluators to independently assess AI risks,” with access “equivalent to that of their own highly privileged employees,” and with no payment “contingent on the evaluator’s findings” (AI Evaluator Forum, letter). In his own comments to the AP he points at the federal agency he used to lead: the labs “aren’t calling for more oversight from the government agency equipped to handle such requests” (AP via PBS).

Gap assessment: No documented gap between statement and act. The structural item is the one his employer states about itself: “Experience shows that privileged-access evaluations inevitably come with downsides” (Transluce, embedded evaluations), and the same employer’s revenue includes the personal holdings of OpenAI and Anthropic employees under a policy that bars funder editorial control. The Forum’s own path-ahead page plans “pooled funding” to fund evaluators “without compromising their impartiality,” which names the funding problem as open. The record does not settle whether the federal agency or the evaluator network is the better home for the standard, and for the draftsman it never needs to.

Interest: (the faction’s stake in each position; a documented fact about the position, never a claim about motive.)

1. Position: embedded third-party evaluators inside every frontier AI company, with employee-equivalent access. The Forum’s letter: “We believe that all frontier AI companies should embed evaluators to independently assess AI risks” (AI Evaluator Forum, letter).
   • What adoption would pay Transluce and the Forum’s members. A rule that labs admit outside evaluators creates the demand the members supply; Transluce’s fundraiser describes revenue from companies and governments, and the Forum lists members that run evaluations. No fee or contract value is published. The AEF-1 page itself concedes: “The success of AEF-1 requires adoption by a range of stakeholders, including frontier developers” (AI Evaluator Forum, AEF-1).
   • His own best case, steelmanned. The letter’s argument: embedded evaluation is credible only with independence, transparency, protection from retaliation and access; “Embedded evaluations cannot address all oversight needs and should be treated as a complement to, rather than a replacement for, broader efforts by frontier AI companies to expand external oversight” (AI Evaluator Forum, letter).
   • Who else gains from the same rule. Every Forum member; the labs themselves, which the letter says some have already called for (“encouraged to see frontier AI companies call for embedding third-party organizations”); and the evaluators competing with them who would meet the same conditions.
2. Position: a published minimum standard for evaluator independence (AEF-1), adopted voluntarily and referenced by regulators. The EU AI Office “has endorsed key provisions of the AEF-1 standard as a means for providers to comply with the independence provisions of the General-Purpose AI Code of Practice” (AI Evaluator Forum, AEF-1).
   • What adoption would pay Transluce. A standard that the evaluators wrote and a regulator accepts sets the bar their own published reports are already built to clear: Transluce’s mental health evaluation is listed as using the AEF-1 checklist (AI Evaluator Forum, AEF-1).
   • His own best case, steelmanned. Independence is opaque without a checklist; the standard asks evaluators to publish a checklist showing how they met a set of operating conditions, so readers can judge the evaluation.
   • Who else gains from the same rule. Labs that wish to cite a recognized independence standard; regulators that want a reference text without commissioning one; METR and SecureBio, listed on the same page as using it.
3. Position: pooled funding for public-interest evaluators. “The Forum will advance mechanisms to fund public-interest evaluators, such as pooled funding, that let them conduct highly rigorous evaluations at the necessary scale and speed, without compromising their impartiality” (AI Evaluator Forum, path ahead).
   • What adoption would pay members. A pooled mechanism would fund the Forum’s own members, Transluce among them; the page names no funder and no amount.
   • His own best case, steelmanned. The letter asks for “funding mechanisms that give them confidence they will remain funded even in these cases,” meaning when findings are unflattering (AI Evaluator Forum, letter).
   • Who else gains from the same rule. Foundations and governments that fund evaluation without wanting to fund one evaluator; the labs, which would stop being the evaluators’ default payer.

Nothing above is a verdict that the positions are wrong. The finding is the alignment of interests: the person who chairs the network of evaluators also chairs the drafting of the rules that decide who counts as an independent evaluator.

Convergent Drive Classification

Self-preservation: The seat’s funding is disclosed and bounded at the employer: no funder editorial control, no paid evaluations, a disclosed split of lab-employee holdings (Transluce, policy). Goal preservation: The goal, defined standards for independent evaluation, is restated across four seats: OMB, the institute, Transluce, the Forum (GovExec bio; Transluce). Resource acquisition: Holds the relationships that decide whose evaluation counts: a federal agency’s history, a lab-facing evaluator’s governance seat, a network of eight organizations (AI Evaluator Forum, members). Self-improvement: The career runs from intelligence-committee staff to the White House AI director’s chair to a standards institute to a nonprofit evaluator’s first Head of Governance, and then to the chair of the trade network (MeriTalk, as reported; Transluce).

Subject is not an AI system. The drives appear anyway — in the head of governance at an evaluator who chairs the forum that writes the evaluators’ rulebook.

Public footprint: Transluce announcement and team page; AI Evaluator Forum; Stanford SAFE Annual Meeting. No personal social account found.

Sources: Transluce Hires a Head of Governance; About — Transluce; Independence and Transparency Policy — Transluce; Some Focus Areas for Embedded Evaluations — Transluce; Help Fund Scalable Democratic Oversight of AI — Transluce; AI Evaluator Forum; AI Evaluator Forum — Members; AI Evaluator Forum — Mission; AEF-1 Standard; AI Evaluator Forum — The Path Ahead; AI Evaluator Forum — Public Letter, Minimum Conditions for Embedding Evaluators; Luma — AI Evaluator Forum; Stanford Center for AI Safety — SAFE Annual Meeting 2026; AP via PBS NewsHour — Anthropic and OpenAI sound the alarm on AI safety (28 Sep 2026); GovExec event bio — Conrad Stosz; MeriTalk — OMB AI Lead Moves to NIST’s Emerging AI Institute; Nextgov — 2023 Rising Star: Conrad Stosz; IBTimes — More Than 100 AI Experts Sign a Letter.