Jack Clark
Last reviewed
May 24, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v1 · 4,055 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 24, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v1 · 4,055 words
Add missing citations, update stale details, or suggest a clearer explanation.
Jack Clark is a British technology writer turned artificial intelligence policy executive who co-founded Anthropic in 2021 and serves as the company's Head of Public Benefit, having previously held the title of Head of Policy.[1][2] Before Anthropic, Clark spent more than four years at OpenAI as its Policy Director, where he managed external communications around the staged release of the GPT-2 language model.[3][4] He began his career as a technology journalist at The Register and Bloomberg News, where he was described by colleagues as the news organisation's "neural network reporter," and since 2016 he has written the weekly "Import AI" newsletter, which is read by roughly 70,000 subscribers in the machine-learning community.[5][6] Clark was a founding steering-committee member of the AI Index at the Stanford Institute for Human-Centered Artificial Intelligence, an inaugural appointee to the United States National Artificial Intelligence Advisory Committee, and one of three speakers invited to brief the United Nations Security Council during its first formal session on artificial intelligence in July 2023.[7][8][9]
Clark was born in 1988 in Brighton, on the south coast of England, and attended Varndean College, a sixth-form college in the city.[6] He went on to read English Literature with Creative Writing at the University of East Anglia in Norwich, graduating with a BA in 2009.[6] His non-technical background, with no formal training in computer science or engineering, is unusual among the founding cohort of the major US frontier-AI laboratories, most of whom hold doctorates in physics, mathematics, or computing. Clark has framed his work in journalism and policy as an effort to make a fast-moving technical field "legible" to a wider non-specialist audience, a stated mission that draws on his literary training.[5]
In multiple public talks Clark has described his teenage interest in science fiction (he has cited authors including Iain M. Banks and William Gibson) as part of his initial draw toward reporting on emerging technology, and his short-fiction writing in the Import AI newsletter borrows themes and stylistic devices from that tradition.[5][6] After university Clark worked briefly as a technology copywriter at Adfero, a London-based content agency, before moving into journalism.[6] His path from English Literature student to AI policy practitioner is sometimes cited in policy-press profiles as an example of the cross-disciplinary backgrounds that have become common in AI governance work.[10]
Clark joined the British technology news website The Register around 2010 as a reporter covering enterprise infrastructure.[6] His beat included distributed systems, data centres, cloud computing, quantum computing, and early machine-learning research, and his self-description as "the world's only distributed systems reporter" became a recurring joke in his later biographical materials.[5][10] During the early 2010s he covered the rise of Hadoop, the consolidation of the hyperscale cloud providers, and the first wave of deep-learning breakthroughs that followed the 2012 ImageNet competition.[11]
In August 2014 Clark left London for San Francisco to join Bloomberg News as a reporter on the enterprise-technology beat.[6] His initial assignment covered Oracle, Hewlett-Packard, Salesforce, and Dell, but he gravitated quickly toward Google and the deep-learning research emerging from companies such as DeepMind, Facebook AI Research, and Google Brain.[10] By 2015 he was filing dispatches for Bloomberg News and Bloomberg Businessweek on convolutional neural networks, recurrent neural networks, and generative models. A profile in the trade publication Talking Biz News later noted that Clark's editor at Bloomberg had initially advised against his eventual move to OpenAI, calling it a poor career decision.[10]
Colleagues described Clark at Bloomberg as a writer who was unusually willing to read primary machine-learning papers and to translate their content for a business readership at a time when most mainstream coverage of artificial intelligence was either dismissive or sensationalist.[5][10]
In 2016, while still working as a journalist, Clark launched Import AI, a free weekly email newsletter that summarises new machine-learning research papers and adds short essays on policy implications and short science-fiction vignettes.[5] The newsletter is published from the domain jack-clark.net and mirrored on Substack at importai.substack.com, and by 2024 it counted roughly 70,000 subscribers, with regular readers including researchers, policy staff, journalists, and corporate strategists.[5][12] Import AI is widely cited by other AI commentators and was one of the earliest dedicated periodicals to provide weekly arXiv coverage to a non-academic audience.[6][12]
The newsletter's format is consistent across issues. Each edition opens with summaries of recent papers, includes shorter items on policy and industry news, and closes with a piece of short fiction (often labelled "Tech Tales") set in a plausible near future. Clark continued publishing Import AI on a weekly cadence through his entire OpenAI tenure and into his time at Anthropic.[5]
Import AI operates on a hybrid Substack model with a free tier covering the weekly research summary and a paid tier that adds early access to longer essays, comments, and additional analytical material.[12] Clark has stated that the newsletter is a personal project written outside his Anthropic duties and that he uses it in part to keep up his own habit of reading new research papers across subfields.[5] Its readership now includes machine-learning researchers at academic and industrial labs, US and UK government policy staff, journalists covering technology and security, and a smaller cohort of speculative-fiction writers who follow it for the Tech Tales section.[12]
Clark joined OpenAI in September 2016, less than a year after the company was founded in December 2015.[4][6] His initial title was Strategy and Communications Director, a role that combined external press relations with policy work.[13] By the time of his June 2019 congressional testimony his title had become Policy Director.[3][4] Clark left Bloomberg with the encouragement of OpenAI's then-leadership and against the advice of his Bloomberg editor.[10]
At OpenAI, Clark sat at the intersection of three functions that were typically separate at older technology companies: communications, policy, and research strategy. He coordinated public-facing announcements about model releases, supervised the policy team that grew up around them, and participated in research projects that examined the social and security implications of AI.[4][14] During his four-year tenure he hired several of the staff who would later lead policy work at other AI laboratories and at policy think tanks, and the OpenAI policy team he built became a template for similar functions at Anthropic, Google DeepMind, and others.[4][27]
In February 2018 Clark co-authored the report "The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation," released as arXiv preprint 1802.07228.[14] The 100-page report was a collaboration among twenty-six researchers from the University of Oxford's Future of Humanity Institute, the University of Cambridge's Centre for the Study of Existential Risk, OpenAI, the Electronic Frontier Foundation, and the Center for a New American Security, among others. Clark was listed as the third author after Miles Brundage and Shahar Avin.[14]
The report enumerated potential misuse scenarios for AI capabilities then on the horizon, including spear-phishing at scale, synthetic media for political disinformation, and the use of off-the-shelf computer vision to direct armed drones. It also proposed four recommendations: closer engagement between AI researchers and policymakers, the development of disclosure norms for dual-use research, the cultivation of responsibility cultures inside AI organisations, and the expansion of public-policy debate beyond a narrow expert community.[14] The paper became one of the most-cited early documents in the AI-misuse literature and is regularly referenced in subsequent academic and government work on AI security risks.
On 14 February 2019, OpenAI announced GPT-2, a 1.5-billion-parameter language model, in a blog post titled "Better Language Models and Their Implications."[15] The announcement broke with then-current research norms by withholding the full trained weights, citing concerns that the model could be used to generate large volumes of plausible synthetic text for disinformation or spam. Clark, as Policy Director, was responsible for managing press inquiries and writing much of the public messaging around the decision.[3][16]
OpenAI released progressively larger versions across 2019: a 124-million-parameter version with the February announcement, a 355-million-parameter version in May, a 774-million-parameter version in August, and the full 1.5-billion-parameter model in November.[16] Clark described this approach to Congress in June 2019 as a "staged release," intended as a prototype for responsible-disclosure norms in machine learning, comparable in his telling to vulnerability-disclosure practices in computer security.[3][17]
The decision attracted criticism from researchers who argued that the model was not in fact dangerous enough to warrant restriction and that OpenAI was using safety as a marketing tactic, as well as praise from policy analysts who welcomed any precedent for caution.[16] Clark defended the policy in interviews with MIT Technology Review and other outlets, framing the staged release as an experiment rather than a permanent commitment, and noting that the underlying scientific community would need to develop better risk-evaluation tools before similar decisions could be made systematically.[16][17] The episode is frequently cited as an early case study in the practice of "responsible release," a concept that later informed Anthropic's Responsible Scaling Policy and the Frontier Model Forum's work on deployment norms.
Clark left OpenAI in December 2020 alongside Dario Amodei, Daniela Amodei, and a small group of researchers who would shortly afterwards found Anthropic.[1][4] In subsequent interviews the departing group cited disagreements about the pace at which OpenAI was scaling its models relative to the development of safety techniques, and concerns about the implications of the 2019 transition from a nonprofit structure to a capped-profit company with a large Microsoft commercial partnership.[1] Clark has spoken publicly about the importance of having a research organisation that places safety at the centre of its work rather than running safety as an internal counterweight to a commercial mission.[9]
Anthropic was incorporated in January 2021 and announced its existence in May 2021 alongside a Series A funding round of approximately 124 million US dollars led by Jaan Tallinn, with participation from James McClave, Dustin Moskovitz, the Center for Emerging Risk Research, and Eric Schmidt.[1] The seven co-founders were Dario Amodei (Chief Executive), Daniela Amodei (President), Jack Clark, Jared Kaplan, Tom Brown, Sam McCandlish, and Chris Olah; an eighth founding employee, Ben Mann, is also frequently described as a co-founder in subsequent company materials.[1][6] Clark held the title Head of Policy from the company's founding through early 2026.[2]
As Head of Policy, Clark led Anthropic's external engagement with US, UK, and European governments, and with multilateral bodies including the OECD, the Global Partnership on AI, and the United Nations.[9][18] He was the principal author or named coordinator of several formal Anthropic policy documents, including written submissions on the EU's EU AI Act, the US National Institute of Standards and Technology's AI Risk Management Framework, and the United Kingdom's proposed AI regulation white paper.[18] Under his direction the policy team at Anthropic grew from a single-person function at incorporation to several dozen staff across Washington, London, San Francisco, and Brussels by 2025, with sub-teams covering research, government relations, standards, and trust and safety.[4][18]
In September 2023 Anthropic published the first version of its Responsible Scaling Policy, a voluntary commitment to define "AI Safety Levels" (modelled on the biosafety-level system) at which the company would be required to implement specific evaluation, containment, and security measures before training or deploying more capable models.[18] Clark was a public spokesperson for the policy and described it as a mechanism to convert abstract safety commitments into operational triggers tied to measurable model capabilities.[18] He has noted in interviews that the RSP framework was influenced by Anthropic's earlier internal work on capability evaluations and by patterns he observed in his coverage of the GPT-2 staged-release debate.[17][18]
On 26 July 2023, Anthropic, Google, Microsoft, and OpenAI launched the Frontier Model Forum, a co-operative industry body intended to share safety best practices and convene technical work on evaluation.[19] Clark represented Anthropic in the formation and governance discussions and continues to participate in Forum working groups.[19] He has also been a regular participant in the AI Safety Summit series, beginning with the first summit at Bletchley Park in November 2023, which produced the Bletchley Declaration on frontier AI risks signed by twenty-eight countries.[18]
Clark has appeared as a witness in front of US legislative bodies on multiple occasions:
| Date | Body | Title at time of testimony |
|---|---|---|
| April 2018 | US House Oversight subcommittee on Information Technology | Strategy and Communications Director, OpenAI[13] |
| June 2019 | US House Committee on Science, Space, and Technology | Policy Director, OpenAI[3] |
| September 2022 | US Senate Committee on Commerce, Science, and Transportation | Co-founder and Head of Policy, Anthropic[20] |
| February 2024 | US House Committee on Science, Space, and Technology | Co-founder and Head of Policy, Anthropic[21] |
| June 2025 | US House Select Committee on the Chinese Communist Party | Co-founder and Head of Policy, Anthropic[22] |
On 18 July 2023, Clark was one of three external speakers, alongside UN Secretary-General António Guterres and Yi Zeng of the Chinese Academy of Sciences, invited to brief the United Nations Security Council during its first formal session on artificial intelligence, held under the United Kingdom's presidency of the council.[9][23] His remarks, later published on his personal website under the title "AI Safety and Corporate Power," argued that governments could not leave the development of advanced AI exclusively to private-sector actors and that robust evaluation systems were a precondition for any sensible regulatory approach.[9]
In April 2022 the United States Department of Commerce appointed Clark as one of the inaugural 27 members of the National Artificial Intelligence Advisory Committee (NAIAC), a body established under the National AI Initiative Act of 2020 to advise the President and the National AI Initiative Office.[7][24] He served on the committee from its first meeting in May 2022 through the conclusion of the first term cycle in April 2025.[7]
On 11 March 2026 Anthropic announced the launch of The Anthropic Institute, a research organisation inside the company that combines its Frontier Red Team, its Societal Impacts team, and its Economic Research team, and that incubates additional teams on AI-progress forecasting and AI in the legal system.[25] Clark was named to lead the Institute and took the new title Head of Public Benefit, replacing his previous title of Head of Policy.[25] The Institute's founding hires included Matt Botvinick on AI and rule of law, Anton Korinek on transformative economics, and Zoë Hitzig on connections between economics and AI development.[25] Clark described the Institute's role as a "research-based information service" intended to help society navigate transitions to more capable AI systems.[25]
The Institute's stated programme spans five research areas: how AI is reshaping the economy and labour markets; how societal resilience can be strengthened against AI-driven disruption; how emerging threats from advanced models can be modelled and mitigated; how AI systems should be aligned with human values; and how governance frameworks should be designed at national and international levels.[25] In its first months the Institute released economic-impact studies, technical evaluation reports, and policy analyses, and Clark and his colleagues gave a series of public talks framing the Institute as Anthropic's vehicle for engaging with social-science and governance research traditionally outside the scope of a frontier-AI laboratory.[25][26]
In a May 2026 Axios "Behind the Curtain" column, Clark stated that he assessed a greater than 60 per cent probability that, by the end of 2028, an AI model would be capable of autonomously training a more capable successor model, a scenario sometimes described in the literature as an "intelligence explosion."[26] He coupled the prediction with calls for stronger evaluation, monitoring, and government-industry coordination, framing the timeline as a reason to accelerate policy work rather than as a forecast to be passively absorbed.[26] The same column reported that the Anthropic Institute's research agenda includes ongoing work on "early signs" of recursive self-improvement, in which AI systems contribute meaningfully to the research and development of their own successors.[26]
In November 2017 Stanford University announced the AI Index, an open, non-profit project under the umbrella of the Stanford Institute for Human-Centered Artificial Intelligence and the broader One Hundred Year Study on Artificial Intelligence.[8] The initiative was conceived by Yoav Shoham, professor emeritus of computer science at Stanford, and assembled by a steering committee that included Ray Perrault of SRI International, Erik Brynjolfsson of the Massachusetts Institute of Technology, and Clark, who at the time was Policy Director at OpenAI.[8] The committee's stated goal was to publish an annual report that would track quantitative measures of AI activity and progress, including research output, hiring, investment, hardware, education, and ethical issues.[8]
Clark served on the AI Index steering committee from its founding in 2017 through 2024, contributing to the annual report's design and to its sections on research, technical performance, and policy.[5][6] The AI Index Report has been published every year since 2017 and is widely cited by governments, academic institutions, and journalists as a reference dataset for trends in the field. Clark's particular contributions, according to his Stanford biographical materials and the report's acknowledgements, included survey design for the policy sections, analysis of research-paper output trends, and review of the technical-performance benchmarking choices.[5][6] He stepped down from the steering committee in 2024 after seven full reporting cycles.[4]
Beyond Import AI, Clark contributes occasional essays to long-form publications, has been a guest on a number of podcasts including the 80,000 Hours podcast and the Weights and Biases Gradient Dissent series, and has spoken at industry conferences such as O'Reilly's AI Conference and Stanford's HAI symposia.[27] He maintains a social-media presence on X (formerly Twitter) under the handle @jackclarkSF.[5]
Clark's prose style in Import AI has acquired a distinctive register that combines technical specificity with science-fiction-influenced imagery. Several of his "Tech Tales" short stories have been collected and reposted by other publications, and his framing devices, such as the phrase "I, for one, welcome our new" used ironically, are widely echoed in machine-learning Twitter discourse.
Clark has been profiled by Channel 4 News, the Financial Times, Axios, and other outlets, generally in the context of Anthropic's policy positions and his own forecasts of AI capability progress.[10][26] In a Channel 4 interview in 2026 he described the potential economic impact of advanced AI as "ten times larger and ten times faster than the industrial revolution," a phrase that circulated widely in subsequent policy commentary.[10] He has consistently argued in interviews and testimony that policy work should focus on capability-triggered safety measures, mandatory evaluations for frontier systems, and international coordination, while declining to endorse specific licensing or pre-clearance regimes.[9][18][21]
In addition to his Anthropic role, Clark holds a non-resident research fellowship at Georgetown University's Center for Security and Emerging Technology, where his work has focused on AI communication, forecasting, and natural-language processing.[27]
Clark is not the recipient of a major academic prize, reflecting his career as a writer and policy practitioner rather than a research scientist. He has, however, been listed by Time magazine and other outlets on annual indexes of influential figures in artificial intelligence, and he is regularly identified by policy-analysis publications as one of the most prominent industry voices in US AI governance debates.[2][10]
In a 2025 announcement Anthropic's seven co-founders, including Clark, jointly pledged to commit 80 per cent of their personal fortunes to charitable causes addressing AI-driven inequality, with the public commitment valued at the time at more than 21 billion US dollars based on Anthropic's then-current secondary-market valuation.[6][10] The pledge has been described in the philanthropic press as one of the largest collective commitments by founders of a single technology company.[10] Clark has not publicly detailed the structure of his individual giving but has indicated that a portion will be directed toward research and advocacy on AI's effects on labour markets and economic distribution.[25]