RUMMAN CHOWDHURY

Behavioral Archetype

THE AUDITOR-ENVOY — Subject built her name as the outside check: the algorithmic-bias researcher who audited Twitter’s own AI, then founded the nonprofit that runs the industry’s public red-teaming events. Then the outside check moved inside two governments at once. In March 2024 the U.S. State Department named her a Science Envoy for Artificial Intelligence. In September 2024 her nonprofit, Humane Intelligence, signed an agreement with the United Nations International Computing Centre to build the UN system’s own AI governance tooling. The organization built to audit power from outside became the vendor two governmental systems separately commissioned to build their AI oversight infrastructure from inside. Both of those seats have since ended or changed hands: her company’s page now describes the envoy post and the DHS board seat in the past tense, and she stepped down as CEO of the nonprofit on 1 August 2025. This profile scores that seam, not the person: the audit method is real, the appointments are real, and the record does not require bad faith on anyone’s part for the structural point to hold.

Essence Indicators

• Current position: founder and CEO, Humane Intelligence PBC. Her company’s page gives her title as “Founder & CEO” and says she “was the first U.S. Science Envoy for Artificial Intelligence, appointed by the Biden administration, and served on the inaugural U.S. Department of Homeland Security AI Safety and Security Board and the City of New York’s AI Steering Committee”; it adds that she “co-founded Humane Intelligence Non-Profit.” Her own site still reads “Appointed as the U.S. Science Envoy for Artificial Intelligence” beside “CEO and founder of Humane Intelligence public benefit corporation” and “Responsible AI Fellow at Harvard University’s Berkman Klein Center for Internet and Society.” The nonprofit’s page says “Mala Kumar was named Executive Director (no more interim!) in April 2026” and lists bias bounties launched on Zindi in 2026 and an open-source red-teaming app due September 2026 (Humane Intelligence PBC; Rumman Chowdhury, About; Humane Intelligence, About).

• Twitter, 2021–2022: Director (subject’s own later bio: Engineering Director) of the Machine Learning Ethics, Transparency and Accountability (META) team, February 2021 to November 2022; led the team’s public audit of Twitter’s image-cropping algorithm and its 2021 study on algorithmic amplification of political content; laid off in the mass layoffs that followed Elon Musk’s takeover

• Accenture, 2017–2021: joined as the company’s first Managing Director building a global Responsible AI practice; developed the “Fairness Tool” with researchers at the Alan Turing Institute; credited with coining the term “moral outsourcing”

• Founded Humane Intelligence: developed the “bias bounty” concept while at Twitter; the organization launched unofficially as “Bias Buccaneers” in 2022 and was incorporated as a nonprofit in May 2023, with Chowdhury as founder and, per her own and the State Department’s description, CEO; she “stepped down as CEO of Humane Intelligence” on 1 August 2025, when Mala Kumar became Interim Executive Director (Humane Intelligence, About), and is “CEO and founder of Humane Intelligence public benefit corporation, and the co-founder of Humane Intelligence, a nonprofit” (Rumman Chowdhury, About); the org’s own account credits it with organizing “the largest ever generative AI red teaming exercise” at DEF CON 31 in 2023

• U.S. Science Envoy for Artificial Intelligence: named by the State Department on March 13, 2024, one of four scientists in that year’s cohort — the first all-female cohort in the program’s history; the Department’s own release describes her as “the CEO of Humane Intelligence, a technology nonprofit that builds a community of practice around evaluations of artificial intelligence (AI) models” and as “a Responsible AI Fellow at Harvard University’s Berkman Klein Center”

• Humane Intelligence–UNICC partnership, September 2024: her nonprofit and the United Nations International Computing Centre announced a partnership “to develop artificial intelligence (AI) governance tools and infrastructure for the efficient and responsible deployment of AI within the UN system,” including a joint “AI Model Assessment Playbook” and pilot assessment tools intended as public goods for UN member states

• Congressional testimony: appeared before a U.S. House hearing on AI policy, “Advancing Innovation in the National Interest” — the same statecraft register (testifying to the legislature that funds the agencies she now represents abroad) documented across other Statecraft Wing profiles in this file

• With her successor. Chowdhury and Mala Kumar, now Executive Director of the Humane Intelligence nonprofit, co-wrote “Why AI evals need to reflect the real world” (Transformer, 10 September 2025): “Leaderboards amplify Goodhart’s Law; that is, when the metric becomes the target, it stops measuring what matters” (Transformer).

• The nonprofit’s FY2024 return names her in an uncorrected excess benefit transaction. Humane Intelligence’s FY2024 Form 990 (filed 26 August 2026), Schedule L Part I, lists “Rumman Chowdhury,” relationship “Former CEO / Officer,” for “Personal rental and furniture expenses paid,” corrected: “No.” Its statement: “Throughout the calendar year 2024, non-compliant apartment rent, personal furniture rentals and other entity insurance payments were made on behalf of and at the direction of former CEO and Board of Director, Rumman Chowdhury, totaling $82,264.92. To-date, the former CEO has paid back the partial amount of $3,314.88. $78,950.04 remains unpaid.” The return says an independent board did not vote to approve the payments, that they were discovered in September 2025, and that the board engaged independent legal counsel in February 2026 (ProPublica, FY2024 Schedule L; ProPublica Nonprofit Explorer). The filing is the organization’s own report; no IRS determination or court finding was located, and no response from Chowdhury was located. See Humane Intelligence.

• Her own company first, then Twitter. In her written testimony she says: “After my time at Accenture, I founded an algorithmic auditing startup, Parity AI, and was approached by Twitter, to build and lead a team focused on Machine Learning Ethics, Transparency and Accountability (META) as its Engineering Director.” Her company’s page says she “founded Parity (acquired by Twitter)”; her bios at UC Berkeley’s Data Edge and the Center for a New American Security call her the former CEO and founder of Parity, “an enterprise algorithmic audit platform company,” and a founding general partner of the Parity Responsible Innovation Fund, “a Responsible Innovation venture capital fund” (House Science Committee, written testimony; Humane Intelligence PBC; Berkeley Data Edge; CNAS).

• House Science Committee, 22 June 2023. The committee’s hearing page lists her among five witnesses, as “Dr. Rumman Chowdhury, Responsible AI Fellow, Harvard University,” with Jason Matheny (RAND), Shahin Farshchi (Lux Capital), Clement Delangue (Hugging Face) and Dewey Murdick (CSET). Her submitted statement makes four recommendations: “support for AI model access to enable independent research and audit,” “investment in, and legal protections for, independent red teaming and ethical hacking,” “the development of a technology oversight body to supplement existing US Government efforts” (she names an “Office of Responsible Use of Technology”) and “development of and participation in a global AI governance organization.” Her line on cost: “Governance is innovation.” On who pays for public red teaming: “Our efforts rely heavily on volunteer effort, corporate or nonprofit funding.” She discloses a seat she holds abroad: “the Center for Data Ethics and Innovation - of which I am a board member” (House Science Committee, hearing page; written testimony).

• The same hearing, Matheny’s answer. Jason Matheny (RAND), asked by Rep. Babin how to square open research with security, said: “before publishing or open sourcing the largest models, I recommend that we require some amount of red teaming and evaluation in order to understand the risk scenarios before we unleash this thing into the world and we’re unable to take it back” (House hearing, 22 Jun 2023).
• DEF CON 31 generative red team, announced 3 May 2023. The AI Village post is signed by Sven Cattell, Austin Carson and Chowdhury. It names the organizers as AI Village with Humane Intelligence, SeedAI and the AI Vulnerability Database, with Scale AI providing the evaluation platform, and the supporters as the White House Office of Science and Technology Policy, the National Science Foundation’s CISE Directorate, the Congressional AI Caucus and the Wilson Center. Models came from Anthropic, Google, Hugging Face, NVIDIA, OpenAI and Stability; Microsoft took part. Her testimony of the next month adds the funder side: the effort “would not be possible without the support of the White House Office of Science, Technology, and Policy, NIST, input from our nonprofit advisors, as well as the major AI companies” (AI Village, 3 May 2023; written testimony). SeedAI joined Partnership on AI as a partner on 7 October 2025, the same release that welcomed the nonprofit she co-founded (Partnership on AI, 7 Oct 2025).

• DHS AI Safety and Security Board, April 2024 (Board terminated effective 7 March 2025). DHS’s November 2024 Roles and Responsibilities Framework for AI in Critical Infrastructure lists “Rumman Chowdhury, Ph.D., CEO, Humane Intelligence” among the Board’s members, beside Sam Altman (OpenAI), Dario Amodei (Anthropic), Satya Nadella (Microsoft), Sundar Pichai (Alphabet), Jensen Huang (NVIDIA), Arvind Krishna (IBM), Lisa Su (AMD), Matt Garman (Amazon Web Services), Fei-Fei Li and Arati Prabhakar, under Secretary Alejandro Mayorkas as chair. The Federal Register notice of 13 March 2025 says the Secretary “terminates these committees under the provisions of the Federal Advisory Committee Act (FACA) of 1972 (5 U.S.C., Ch.10), as well as 6 U.S.C. 451 effective March 7, 2025,” the Artificial Intelligence Safety and Security Board among them. DHS says the Board “advises the Secretary, the critical infrastructure community, other private sector stakeholders, and the broader public,” that “The duties of the Board are solely advisory in nature” and that “Board Members serve as uncompensated representatives of their sectors.” The Framework the Board was consulted on “Proposes a set of voluntary responsibilities” for cloud providers, AI developers and infrastructure operators. Her company’s page lists the seat and a second one, on the City of New York’s AI Steering Committee (DHS, Framework, November 2024; Humane Intelligence PBC; Federal Register, 2025-04011).

• A NIST-supported red-team exercise, August to October 2024. On 21 August 2024 she sent the Bias Bounty board list an announcement, forwarded to a Carnegie Mellon list the next day, that the nonprofit was launching a nationwide online qualifier, “a NIST ARIA (Assessing Risks and Impacts of AI) pilot exercise,” with the in-person event “to be held alongside CAMLIS, an AI security conference” and described as “supported by the U.S. National Institute of Standards And Technology”; the email says “All expenses for travel, food and lodging during this time will be covered” (forwarded announcement, Carnegie Mellon Auton Lab list, 22 Aug 2024). NIST’s current consortium roster lists Google, OpenAI, Anthropic, MLCommons and Partnership on AI; it does not list Humane Intelligence (NIST AI Consortium Members, updated 20 October 2026).

• Stanford’s transparency index. The Foundation Model Transparency Index’s advisory board lists Arvind Narayanan, Daniel E. Ho, Danielle Allen, Daron Acemoglu and Chowdhury; the index’s 2025 edition scores thirteen companies (Stanford CRFM, FMTI December 2025; see Stanford CRFM).

• On evaluator independence, July 2026. In Fathom’s “Who Evaluates the Evaluators?” (21 July 2026), identified as “founder and CEO of Humane Intelligence PBC,” she says: “A hundred percent of the time, I am co-working to help define good,” and, of her firm’s role, “We help improve discoverability, and match people to the tools that solve their problem” (Fathom, 21 Jul 2026).

• Her company’s client list. Humane Intelligence PBC’s page, under “Trusted by enterprise & institutional partners,” shows General Motors, Autodesk, American Family Insurance, Biogen and DNB; it sells “Evaluation-as-a-Service” and a “Platform-Only” tier, and names no investor, funder or government client (Humane Intelligence PBC).

Social Persona / Impression Management

Immediate impression: The credentialed critic. A decade of TED talks, magazine profiles, and “shaping AI” lists built the persona of the outside technologist who makes power show its work — brisk, fluent in both the statistics and the press cycle.

Energy: Method-first, infrastructure-first. She does not argue a single model’s output. She builds the tool — the Fairness Tool, the bias bounty, the red-team event format — and then the organization that runs the tool at scale.

Impression management strategy: The independent auditor. The framing is durable because it was true at each individual step: an internal ethics team that got real audits published over engagement-team objections, a nonprofit that ran the largest public AI red-teaming event on record. Her own words from February 2023, three months after the layoff, stated the thesis plainly: “I saw firsthand the harms that can flow from unchecked power in tech.” What the later record adds is that “power” now includes the U.S. State Department, which appointed her, and the United Nations’ own computing arm, which hired her nonprofit to build its AI governance tooling. The auditor’s seat and the audited institution’s seat are occupied by the same desk. The record does not show which one is operating at any given moment, and for the position it does not need to.

• Social accounts and feeds. Her own site, rummanchowdhury.com/about, links Bluesky @ruchowdh.bsky.social, Instagram rumman_c and LinkedIn (not ground truth); it links no X account, and its newsletter form was not working when read. Her site and her company’s page offer no blog feed (the feed addresses return 404). The nonprofit she co-founded publishes an organization feed at humane-intelligence.org/feed and a Bluesky account, which are the organization’s, not hers. Her essays run on others’ pages: The Atlantic (2023), Transformer (2025, with Mala Kumar) and an interview on Fathom (2026).

Forensic Archetype Comparison

Psychometric Assessment

Big Five (OCEAN):

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

MBTI: ENTJ-adjacent — builds systems (tools, organizations, frameworks) to operationalize a stated principle, then moves the system to wherever it can operate at the largest scale available.

Threat Assessment

Alliance Map

These run at the same time. In 2024 alone she held a State Department envoy appointment (March), a seat beside the chief executives of OpenAI, Anthropic, Microsoft, Alphabet, NVIDIA and IBM on a DHS advisory board (April), a UN agreement through the nonprofit (September) and a NIST-supported red-team exercise (August to October), while the company she founded sold evaluation to private clients and the nonprofit took grant money from Google.org, whose parent’s models it had tested at DEF CON. Each tie is stated by her, by the counterparty or by a government document. The overlap is the finding; the record does not show which seat she acts from on any given day.

Alignment Analysis

Stated alignment: Build independent, community-driven AI auditing so that no single company or government owns the definition of “safe AI”; expand public access to red-teaming and bias-testing tools; hold power accountable through external, outside-the-institution review — the case she made in her own words after leaving Twitter: legal mandates for source-code audits, bias bounties as standard practice, “genuine oversight.”

Observed alignment: The audit infrastructure she built was delivered through two government-adjacent contracts at once — a U.S. State Department diplomatic appointment, and a signed agreement to build the United Nations’ own AI Model Assessment Playbook for use inside the UN system.

Gap assessment: The stated and observed alignments overlap wherever “independent audit access for AI oversight” coincides with “AI oversight infrastructure hosted and commissioned by the government and the intergovernmental body being overseen” — and the current arrangement puts the overlap on the table rather than hiding it. The 2023 argument was explicitly for outsiders to hold institutional power accountable from outside; the 2024 record shows the same organization now on the inside of two of the largest institutional AI-governance efforts in the world. The record does not establish that this makes either the State Department portfolio or the UNICC playbook less rigorous — only that “the independent check” and “the commissioned infrastructure” are, at present, performed by the same desk, and the position does not require the record to settle which one is operating at any given moment. Both seats have since ended or changed hands, and the desk now runs a company that sells evaluation to private clients.

Interest: Her 2023 testimony asks for four things, and the first two describe the market she sells into: “funding for independent groups to conduct red teaming and adversarial auditing,” legal protection so such researchers “are not silenced with litigation,” and model access so that “impartial and objective external actors” can “verify what the true impact of these models are.” What adoption would pay her: the nonprofit sells red teaming “as a paid service” and contextual evaluations, her company sells “Evaluation-as-a-Service” and “Platform-Only” access, and her testimony notes that “The few individuals who are skilled at these tasks are often hired by tech companies.” Neither entity’s page says it has applied for, or been named in, any statutory designation. Her best case, in her words: “In my years of experience delivering industry solutions in Responsible AI, good governance practices have contributed to more innovative products,” and “the real competitive advantage is trustworthiness.” Others who gain from the same rules: every evaluator that sells testing, and the labs, which supplied the models for the public event and are named in her testimony as supporters (written testimony; Humane Intelligence PBC; Humane Intelligence).

Convergent Drive Classification

Self-preservation: Carries one method — build the audit, then the organization that runs it — across Twitter, Accenture, her own nonprofit, a national diplomatic portfolio, and a UN system contract. Goal preservation: Named the terms first (Fairness Tool, bias bounty, the DEF CON red-team format), so whichever institution hires the organization, the audit runs on her framework. Resource acquisition: Holds the scarcest resource in AI governance right now — a nonprofit trusted enough that a great power’s State Department and the United Nations’ own computing arm each separately commissioned it to build their AI oversight infrastructure. Self-improvement: Platform ethics director, then consultancy practice lead, then nonprofit founder, then national envoy plus IGO contractor — the same instinct, applied at a higher institutional altitude each time.

Subject is not an AI system. The drives appear anyway — in the auditor who became two governments’ audit infrastructure.

Sources: Announcement of the 2024 Cohort of U.S. Science Envoys — U.S. Department of State (Mar 13, 2024); Rumman Chowdhury — Wikipedia; About — Humane Intelligence; About — rummanchowdhury.com; UNICC and Humane Intelligence to Develop AI Governance Tools and Infrastructure — UNICC (Sept 22, 2024); I Watched Elon Musk Kill Twitter’s Culture From the Inside — The Atlantic (Feb 17, 2023); Transcript: House Hearing on AI - Advancing Innovation in the National Interest — Tech Policy Press; Humane Intelligence PBC; Notice of Termination of Discretionary Federal Advisory Committees — Federal Register (13 Mar 2025); Form 990 Schedule L — ProPublica Nonprofit Explorer.