PATRICIA PASKOV

Behavioral Archetype

THE STANDARDS DIRECTOR WITH A SEAT AT EACH TABLE — Subject is the person who leads the writing of the audit standards at the evaluator that advocates mandatory audits, who is also Senior Advisor to the Forum of eight evaluators that issued the first operating-conditions standard, and who holds a RAND affiliation, an Oxford Martin affiliation, a lead role in an intergovernmental AI-safety report and a mentor slot at MATS. AVERI’s team page: “Patricia Paskov is the Standards Director at AVERI, where she leads the development of standards for frontier AI auditing.” Her MATS stream lists as sample projects “Expansion and governance strategy for AI Evaluators Forum” and “Building out a PCAOB for AI auditing and/or FINRA for AI models.” The hazard is not a gate. She approves, funds and blocks nothing. It is that one person’s name sits on the standard, on the Forum that adopted it, on the organization that lobbies for the audit requirement the standard serves, and on the research institutions whose pages describe her. This profile scores the position, not the person.

Essence Indicators

• Standards Director, AVERI. AVERI’s team page: “Patricia Paskov is the Standards Director at AVERI, where she leads the development of standards for frontier AI auditing. Patricia is also an Adjunct Researcher at RAND, a Research Affiliate at the Oxford Martin AI Governance Initiative, and Faculty at the International Programme on AI Evaluation. She led the Resilience section of the 2026 International AI Safety Report, and her research has appeared in top AI/ML venues and been covered by outlets including Al Jazeera, the Financial Times, Fortune, MIT Technology Review China, and TIME.” It adds that she “built machine learning models at Condé Nast and designed large-scale randomized controlled trials with the World Bank and Innovations for Poverty Action” and “is completing a DPhil in Engineering Science at the University of Oxford.” AVERI’s about page: “We are a standard-setter grounded in practice,” and “We lead the development of frontier AI auditing standards, or the criteria for what counts as good audits and qualified auditors.” It also 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” (AVERI, Team; AVERI, About). See AVERI and Miles Brundage.

• Senior Advisor, AI Evaluator Forum. The Forum’s members page lists under “Leadership” “Conrad Stosz, Chair” and “Patricia Paskov, Senior Advisor,” with no affiliation printed beside either. Its AEF-1 standard, “Minimum Operating Conditions for Independent Third Party AI Evaluations,” “Version 1, updated December 4, 2025,” lists her ninth of 26 authors with affiliation 5, “Oxford Martin AI Governance Initiative,” while Brundage, Grace Werner, George Balston and Sean McGregor carry affiliation 3, “AI Verification and Evaluation Research Institute.” Brundage writes that AVERI “is a founding member of the AI Evaluator Forum and some AVERI staff helped write their first standard, AEF-1” (AI Evaluator Forum, Members; AI Evaluator Forum, AEF-1; Miles Brundage, The Launch of AVERI). See AI Evaluator Forum, Conrad Stosz and Grace Werner.

• Author, ‘Frontier AI Auditing’ (16 January 2026). She is the fifth of 48 listed authors on arXiv, after Brundage, Noemi Dreksler, Aidan Homewood and Sean McGregor and before Stosz (sixth). The paper’s note says inclusion “does not entail endorsement of all claims in the paper, nor does authorship imply an endorsement on the part of any individual’s organization” (arXiv:2601.11699; arXiv HTML, v1).

• Contributor, International AI Safety Report 2026 (24 February 2026). The report’s arXiv listing names “Patricia Paskov” among its authors (“Yoshua Bengio and 91 other authors”). Her Oxford Martin page and RAND page both say she is “the Resilience section lead for the 2026 International AI Safety Report” (arXiv:2602.21012; Oxford Martin AIGI, Patricia Paskov; RAND, Patricia Paskov).

• RAND and Oxford Martin. RAND’s page: “Patricia Paskov is an adjunct researcher at RAND focused on frontier AI capability evaluations and policy. She helped launch the RAND Center on AI, Security, and Technology’s EU AI policy workstream, and currently researches the science of evaluations, open questions in human uplift studies, and AI autograders. She is also the Resilience section lead for the 2026 International AI Safety Report and a Working Group Chair at the EvalEval Coalition.” Her Oxford Martin page: “Patricia Paskov is a Research Affiliate of the Oxford Martin AI Governance Initiative and a Technology and Security Policy Fellow at RAND,” lists “Open Problems in Frontier AI Risk Management” (24 February 2026), “Safety Frameworks and Standards: A comparative analysis to advance risk management of frontier AI” (9 October 2025) and “In Which Areas of Technical AI Safety Could Geopolitical Rivals Cooperate?” (23 May 2025), and says her ICML work “received a 2025 Spotlight Award” and that she “has presented her findings to key stakeholders including the US Department of Defense, the EU AI Office, and Microsoft Research.” AEF-1’s reference list includes “Patricia Paskov, Michael J. Byun, Kevin Wei, and Toby Webster. Preliminary suggestions for rigorous GPAI model evaluations. Technical report, May 2025” (RAND, Patricia Paskov; Oxford Martin AIGI, Patricia Paskov; AEF-1). See RAND.

• Mentor at MATS, Winter 2027. MATS’s page for the stream “Patricia Paskov & Miles Brundage (AVERI)” says it “will focus on AI auditing, evaluation standards, and governance,” lists the sample projects “Expansion and governance strategy for AI Evaluators Forum,” “Standards for, e.g., access, auditing governance, and conflict of interest,” “Building out a PCAOB for AI auditing and/or FINRA for AI models” and “Ecosystem intelligence for the state of AI auditing, potentially resulting in a living repository or recurring flagship report on: auditing regulations, ISO and standards, licensing and credentials, auditing orgs, and beyond,” and her bio adds that she “co-founded NYC AI Governance & Safety, and serves as Faculty at the International Programme on AI Evaluation” (MATS, Patricia Paskov & Miles Brundage (AVERI); MATS, Patricia Paskov (Trish)). See MATS.

• In her words, on method (Cambridge CHIA seminar, 13 March). The seminar’s title is “RCTs Meet Frontier AI: Why Human Uplift Studies Are Hard and What We Can Do About It.” The abstract: “Human uplift studies — studies that measure AI’s effects on human performance using randomized controlled trial methodology — are increasingly used to inform deployment, governance, and safety decisions for frontier AI systems.” And: “rapidly evolving models, shifting baselines, heterogeneous user proficiency, and porous real-world settings all strain core assumptions of causal inference.” It rests on “interviews with 16 expert practitioners across biosecurity, cybersecurity, education, and labor” (CHIA, CHIA Seminar Series – Patricia Paskov).

• Social accounts. MATS’s mentor profile for her links x.com/prpaskov. AVERI’s, RAND’s, Oxford Martin’s and Foresight’s pages link no personal X account, and no post is quoted here. LinkedIn is linked from RAND’s page and is not used as a source.

Social Persona / Impression Management

Immediate impression: The method-minded standards writer. An economist who ran randomized trials for the World Bank, built models at a media company, and now writes the criteria for audits of frontier AI developers. The footprint is five institutional bios, three co-authored documents, a conference award and a seminar abstract.

Energy: Convening and credentialing. Her public record is bios that stack roles, a mentor page, a seminar abstract and author lists; the record read contains no op-ed, interview transcript or post in her own voice.

Impression management strategy: The multiple-affiliation bio: each institution’s page repeats the others’ titles. AVERI’s lists RAND, Oxford and the intergovernmental report; Oxford’s lists RAND and the report; RAND’s lists the report and EvalEval; MATS’s lists all of them (AVERI, Team; Oxford Martin AIGI; RAND; MATS).

First-hand exhibits. Her only extended words are the CHIA abstract and the sample-project lines on MATS’s stream page.

Forensic Archetype Comparison

Psychometric Assessment

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

Big Five (OCEAN) (inferred from the public role and public record only):

Dark Triad (evidence-bound):

MBTI: INTJ-adjacent (“The Architect”), inferred from the documented public role only. The rationale is the work product: standards, frameworks and criteria. First-hand rationale: “She leads the development of standards for frontier AI auditing” (AVERI, Team). A style inference, not a test result.

Threat Assessment

Score rationale. The receipts: AVERI’s team and about pages, the Forum’s members page and AEF-1, the arXiv auditing paper and report listing, RAND’s, Oxford Martin’s, Cambridge’s and MATS’s pages. Attack 5, defense 4, durability 4, hazard 43.3. Peers: Conrad Stosz (5/5/5, hazard 40), the Forum’s Chair: level on attack, since both write the criteria and the Forum’s own page names both; a point below on defense and durability because the Chair seat is the named office and she is the advisor beside it. Kai Fronsdal (4/3/4, hazard 36.7): a point below on attack, because his leverage is a tool inside others’ evaluations and hers is the standard those evaluations are held to, and a point below on defense, because he holds one staff post and she holds several affiliations. Alexandra Abbas (3/3/4, hazard 33.3), a program lead at Meridian: two below on attack and one below on defense for the same reasons. Chris Painter (6/8/7, hazard 62), the government-facing evidence broker: a point above on attack because his work reaches governments directly, four and three points above on defense and durability because his institution is a standing evaluator he helps run. Attack 5 is the number most open to challenge: the pages state leading a development effort, and the effect of the standards on any decision is not documented.

Score basis. ATK 5 is leverage: leads standards development at AVERI, Senior Advisor to the Forum, author of AEF-1 and the auditing paper, contributor to the 2026 report (AVERI, Team; AI Evaluator Forum, Members). DEF 4 is insulation: a staff post and four concurrent affiliations, none with a term, none the sole source of standing (Oxford Martin AIGI). HP 4 is durability: cited work that persists, with no standing institution of her own (AEF-1).

Alignment Analysis

Stated alignment: Standards and rigor for evaluating and auditing frontier AI. AVERI: “We are a standard-setter grounded in practice” (AVERI, About); her seminar abstract on method: the challenges “strain core assumptions of causal inference” (CHIA). Observed alignment: Consistent. She leads the writing of audit criteria at the audit advocate, advises the Forum that issued the operating-conditions standard and sits in two research institutions that write on the same questions (AVERI, Team). Gap assessment: None documented. The stream’s own sample projects include “Standards for, e.g., … conflict of interest” (MATS). The record cannot settle which of her four affiliations pays her, what her AVERI role’s start date is, or why the Forum’s AEF-1 PDF puts her at Oxford Martin while AVERI’s team page puts her at AVERI; none of the pages says.

Interest: Documented, never motive, never a verdict. She has published no policy argument outside co-authored documents and an abstract; the positions below are the ones those documents and her mentor page carry.

1. Position: audit standards, including access, auditing governance and conflict of interest, should be written and then given an oversight body (her MATS stream’s sample projects: “Standards for, e.g., access, auditing governance, and conflict of interest” and “Building out a PCAOB for AI auditing and/or FINRA for AI models,” MATS). What adoption pays: AVERI is the organization that says “We lead the development of frontier AI auditing standards,” helped write the Forum’s first standard and endorsed Illinois SB 315; she is its Standards Director and the Forum’s Senior Advisor, and the regime her stream describes would need standard-setters (AVERI, About). No fee, grant or contract value for her is published. Best case: AVERI’s: audits need “deep, secure access to non-public information – access that developers aren’t currently required to provide.” Others who gain: the member evaluators, qualified auditors and the licensing bodies a regime creates.
2. Position: evaluation evidence has validity limits that standards should state (CHIA abstract: uplift studies are “increasingly used to inform deployment, governance, and safety decisions,” while “porous real-world settings all strain core assumptions of causal inference,” CHIA). What adoption pays: a method-first standard widens the market for evaluators able to meet it, among them the RAND and Oxford groups she belongs to (RAND). Best case: her abstract’s: “clarifying both the limits and appropriate uses of uplift evidence in high-stakes decision-making.” Others who gain: labs and regulators relying on uplift evidence, which gain a clearer account of what it can bear.

Alliance Map

The Corporate and Movement rows overlap in people and in time. The organization that writes the audit standards, the Forum of evaluators that adopted the first one, the research institutions that publish on the same questions and the program that trains the next cohort all list the same person, in the open and on the record. The overlap is the finding, not a coordination claim. See the small world.

Convergent Drive Classification

Self-preservation: Holds several affiliations at once, so that no single seat is load-bearing. First-hand: “Adjunct Researcher at RAND, a Research Affiliate at the Oxford Martin AI Governance Initiative, and Faculty at the International Programme on AI Evaluation” (AVERI, Team). Goal preservation: Keeps one object, the audit criteria, across the advocate, the Forum and the mentor pipeline. First-hand: “Expansion and governance strategy for AI Evaluators Forum” (MATS). Resource acquisition: No compensation or grant is documented for her. The documented resources are institutional: a research affiliation, a standard-writing mandate and a mentor stream (RAND). Self-improvement: From economics and randomized trials to ML to evaluation science to standards. First-hand: “completing a DPhil in Engineering Science at the University of Oxford” (AVERI, Team).

Subject is not an AI system. The drives appear anyway — in the standards director whose name is on every page that describes the criteria.

Public footprint: AVERI team · RAND profile · Oxford Martin AIGI profile.

Sources: AVERI, Team; AVERI, About; AI Evaluator Forum, Members; AI Evaluator Forum, AEF-1; arXiv:2601.11699, Frontier AI Auditing; arXiv HTML v1; arXiv:2602.21012, International AI Safety Report 2026; RAND, Patricia Paskov; Oxford Martin AIGI, Patricia Paskov; CHIA, CHIA Seminar Series – Patricia Paskov; MATS, Patricia Paskov (Trish); MATS, Patricia Paskov & Miles Brundage (AVERI); Miles Brundage, The Launch of AVERI.