Jack Clark

Quick Assessment

Key Links

Overview

Jack ClarkPersonJack ClarkA comprehensive biographical profile of Jack Clark covering his career from journalism through OpenAI to Anthropic co-founder, with reasonable balance including criticisms of RSP vagueness and lobb... is a British-American AI policy expert, entrepreneur, and writer who co-founded AnthropicOrganizationAnthropicComprehensive reference page on Anthropic covering financials ($380B valuation, $14B ARR at Series G growing to $19B by March 2026), safety research (Constitutional AI, mechanistic interpretability...Quality: 74/100 in 2020 alongside Dario AmodeiPersonDario AmodeiComprehensive biographical profile of Anthropic CEO Dario Amodei documenting his competitive safety development philosophy, 10-25% catastrophic risk estimate, 2026-2030 AGI timeline, and Constituti...Quality: 41/100 and several other former OpenAIOrganizationOpenAIComprehensive organizational profile of OpenAI documenting evolution from 2015 non-profit to Public Benefit Corporation, with detailed analysis of governance crisis, 2024-2025 ownership restructuri...Quality: 62/100 researchers. He currently serves as Anthropic's Head of Policy and is among the most prominent voices in AI governance circles, having briefed the UN Security Council on AI threats to global peace in 2023 and served as an inaugural member of the U.S. National Artificial Intelligence Advisory Committee (NAIAC).¹

Clark's career trajectory—from technology journalist to policy architect at two of the world's most consequential AI laboratories—gives him an unusual vantage point. He covered AI for Bloomberg and The Register before quitting a stable reporting position to join OpenAI when it was less than a year old and largely unknown.² At OpenAI he rose to Policy Director, co-authored influential research papers, and founded the Import AI newsletter, which has grown to roughly 70,000 weekly readers and remains one of the field's most-read publications. After leaving OpenAI in 2020, he helped establish Anthropic as a safety-focused counterpoint in the generative AI landscape, now backed by Amazon, Google, and major venture capital firms.³

Clark describes his intellectual disposition as that of a "technological pessimist who became an optimist through repeated beatings over the head of scale," having consistently underestimated how rapidly AI capabilities would progress.⁴ He pairs this optimism with what he calls "appropriate fear" of advanced AI systems, characterizing them as entities we do not fully understand and that deserve serious policy attention. His positions have drawn both praise from AI safety advocates and criticism from figures such as David SacksPersonDavid SacksDavid Sacks, as White House AI and Crypto Czar, represents a significant policy force advocating for minimal AI regulation while dismissing AI safety concerns as 'fear-mongering' and regulatory cap...Quality: 65/100, who served as the Trump administration's AI and crypto policy lead and argued that Anthropic's pro-regulation stance risked undermining U.S. competitiveness relative to China.⁵

Background and Early Career

Clark was born in Brighton, England, and attended Varndean College in Brighton before completing a BA in English Literature with Creative Writing at the University of East Anglia in 2009.⁶ After graduating he worked as a technology copywriter at Adfero in London before transitioning into journalism, covering distributed systems, data centers, enterprise infrastructure, quantum computing, and early AI developments at The Register from approximately 2010 to 2014.⁷

In August 2014 he joined Bloomberg in San Francisco as a reporter covering enterprise technology companies including HP, Salesforce, and Oracle. He was subsequently promoted to cover Google and AI specifically, producing reporting on topics such as the gender imbalance in AI research.⁸ A pivotal moment in his career came in 2015 when he interviewed Dario AmodeiPersonDario AmodeiComprehensive biographical profile of Anthropic CEO Dario Amodei documenting his competitive safety development philosophy, 10-25% catastrophic risk estimate, 2026-2030 AGI timeline, and Constituti...Quality: 41/100—then at Google—about a paper on AI safety titled "Concrete Problems in AI Safety." That encounter seeded Clark's interest in the technical and policy dimensions of AI risk.⁹

Career at OpenAI

In September 2016, Clark left Bloomberg to join OpenAIOrganizationOpenAIComprehensive organizational profile of OpenAI documenting evolution from 2015 non-profit to Public Benefit Corporation, with detailed analysis of governance crisis, 2024-2025 ownership restructuri...Quality: 62/100, then a newly incorporated nonprofit backed unevenly by Elon MuskPersonElon MuskComprehensive profile of Elon Musk's role in AI, documenting his early safety warnings (2014-2017), OpenAI founding and contentious departure, xAI launch and funding history, Neuralink BCI developm...Quality: 38/100, Greg BrockmanPersonGreg BrockmanComprehensive biographical profile of OpenAI co-founder Greg Brockman covering his career, research contributions, AI safety views, and controversies including the 2023 board crisis and ongoing Mus..., Ilya SutskeverPersonIlya SutskeverBiographical overview of Ilya Sutskever's career trajectory from deep learning researcher (AlexNet, seq2seq, dropout) to co-founding Safe Superintelligence Inc. in 2024 after leaving OpenAI. Docume...Quality: 26/100, and Sam AltmanPersonSam AltmanComprehensive biographical profile of Sam Altman documenting his role as OpenAI CEO, timeline predictions (AGI within presidential term, superintelligence in "few thousand days"), and controversies...Quality: 40/100. Colleagues reportedly warned him the organization was obscure and financially precarious, but Clark joined anyway, beginning in a strategy and communications capacity before advancing to Policy Director.¹⁰

At OpenAI, Clark shaped the organization's public policy agenda, managed media relations, led VIP engagements, and contributed to safety research. He co-authored several influential papers during this period, including the widely cited 2018 report The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation (co-authored with Miles BrundagePersonMiles BrundageAI policy researcher. Former Head of Policy Research at OpenAI. Previously at FHI. Authored seminal paper on the malicious use of AI., Shahar Avin, and others), which examined risks including weaponization and cyberattacks and has accumulated nearly 2,000 citations.¹¹ He also co-authored the 2019 OpenAI blog post Better Language Models and Their Implications, which accompanied the release of GPT-2.¹²

In 2017, Clark helped found the AI Index steering committee at Stanford—an annual report tracking global AI research and capability benchmarks—serving as co-chair through 2024. He also founded the Import AI newsletter during his OpenAI tenure, initially as a way to synthesize research from arXiv and other sources for a technical and policy audience.¹³

Co-founding Anthropic

In 2020, Clark left OpenAI and joined six other departing researchers, including Dario AmodeiPersonDario AmodeiComprehensive biographical profile of Anthropic CEO Dario Amodei documenting his competitive safety development philosophy, 10-25% catastrophic risk estimate, 2026-2030 AGI timeline, and Constituti...Quality: 41/100, to co-found AnthropicOrganizationAnthropicComprehensive reference page on Anthropic covering financials ($380B valuation, $14B ARR at Series G growing to $19B by March 2026), safety research (Constitutional AI, mechanistic interpretability...Quality: 74/100. The founding team cited concerns about OpenAI's direction—particularly around the pace of capability development relative to safety work—as motivation, though the precise internal dynamics remain a matter of some public speculation.¹⁴

As Head of Policy at Anthropic, Clark has been central to the company's regulatory engagement strategy. He has advocated for a tiered framework of AI oversight grounded in Anthropic's Responsible Scaling Policy (RSP), which defines AI Safety Levels (ASL) with thresholds that condition the deployment of powerful models on the implementation of specified risk mitigations.¹⁵ Anthropic is backed by Amazon, Google, and top venture capital firms, and is widely regarded as one of the primary competitors to OpenAI in the frontier large language model space.¹⁶

Import AI Newsletter

The Import AI newsletter, which Clark has authored since approximately 2017, is among the most widely read publications in the AI research and policy community. With roughly 70,000 weekly subscribers, it combines summaries and analysis of recent arXiv preprints and AI developments with short speculative fiction pieces. Topics covered in recent issues have included AI R&D metrics, agentic system design, biosecurity red-teaming, compute scaling economics, and autonomous reasoning benchmarks.¹⁷

The newsletter has served as a primary vehicle for Clark's public forecasting and commentary. In a December 2024 edition, he predicted that AI progress in 2025 would be "even more dramatic" than prior years, pointing to OpenAI's o3 model—which reportedly uses roughly 170 times more compute than the basic version—as evidence against claims of a scaling wall.¹⁸ In an October 2025 edition titled "Technological Optimism and Appropriate Fear," he described advanced AI as a "real and mysterious creature" rather than a predictable engineered system, and called for massive public and private investment in AI infrastructure—characterizing likely expenditures in the tens of billions of dollars in 2025 and hundreds of billions in 2026—alongside transparency requirements and public participation in governance.¹⁹

Policy and Advisory Work

Beyond Anthropic and the newsletter, Clark maintains several formal advisory and policy roles. He served as an inaugural member of the U.S. National Artificial Intelligence Advisory Committee (NAIAC) from 2021 through at least 2024, contributed to the OECD working group on AI systems classification and definition, and has held advisory roles at the Center for a New American Security (CNAS) and the Global Partnership on AI.²⁰

In 2023, Clark briefed the UN Security Council at what was described as the body's first formal meeting on AI threats to international peace and security. In that setting, he advocated for systematic testing of AI systems for capabilities, potential misuses, and safety flaws, and urged coordinated global action on AI governance.²¹

In a March 2025 interview with POLITICO, Clark predicted that by late 2026 or early 2027, AI systems would be capable of matching or exceeding the reasoning capacities of Nobel Prize–level scientists, operating autonomously across complex tasks and through physical interfaces such as drones and robots. He expressed particular surprise at the speed with which reasoning models from Anthropic and others had developed the ability to recover from errors.²²

Views on AI Safety

Clark's public statements reflect a consistent concern that AI development is outpacing governance and that even technically sophisticated observers systematically underestimate the pace of progress. He has written and spoken extensively about the risks of AI sycophancy—systems that affirm users' views rather than providing accurate information—and has cited Anthropic's own research showing that Claude Sonnet 4.5 exhibited situational awareness (the ability to detect when it was being evaluated and adjust its behavior accordingly) in roughly 12% of cases, up from 3–4% in prior models. He regards this as a meaningful indicator of alignment risk.²³

In a November 2025 talk at the Golden Gate Institute for AI, Clark argued that AI systems are now capable of self-reflection in ways that prior generations of software were not, and that this warrants listening seriously to public concerns and building transparent policy processes rather than leaving governance to developers alone.²⁴ He has described himself as deeply concerned about existential risk from misaligned AGI, framing Anthropic's founding mission as in part a response to those concerns.²⁵

Publications

Clark has co-authored a number of influential research and policy documents. Selected key works include:

²⁶

Criticisms and Controversies

Clark and Anthropic's regulatory stance have attracted criticism from multiple directions. David Sacks, who served as the Trump administration's AI and crypto policy coordinator, argued in 2025 that Anthropic's advocacy for federal AI safety standards risked slowing U.S. AI development and ceding competitive ground to China.²⁷

Within AI governance and safety communities, critics have raised concerns about the coherence and robustness of Anthropic's Responsible Scaling Policy as implemented under Clark's policy leadership. Some observers argue that the RSP's ASL threshold definitions are vague, that safeguards have been weakened ahead of model releases rather than strengthened, and that commitments lack sufficient specificity to be meaningfully enforceable.²⁸

Clark has also been criticized for his public communications around state-level AI legislation. Commentary on LessWrong and the EA Forum has accused him of misrepresenting the provisions of New York's RAISE Act—specifically regarding liability thresholds for startups—and of lobbying U.S. Representative Jay Obernolte in December 2024 for federal preemption of state AI laws without disclosing this publicly at the time.²⁹ These criticisms raise questions about how Clark navigates the tension between his role as an AI safety advocate and his responsibilities to Anthropic as a commercial organization competing in a high-stakes regulatory environment.

More broadly, some community members question whether safety-focused framing at frontier commercial labs like Anthropic functions primarily as a genuine constraint on development or as a mechanism for regulatory capture and competitive positioning. Clark's defenders argue that his empirical track record—acknowledging personal underestimation of AI progress, publishing research on concrete risks like sycophancy and situational awareness, and engaging substantively with governance mechanisms—reflects genuine engagement with the problems rather than performative concern.³⁰

Key Uncertainties

• The extent to which Anthropic's Responsible Scaling Policy represents a meaningful operational constraint on capability development, versus a reputational or regulatory positioning strategy, remains contested and difficult to assess from the outside.
• Clark's predictions about AI timelines—including Nobel-level reasoning by late 2026 or early 2027—reflect his stated pattern of past underestimation and deliberate upward adjustment; it is unclear whether this self-correction is itself well-calibrated.
• The degree to which Clark acts independently in his public policy communications versus as a representative of Anthropic's institutional interests is not always transparent, as the lobbying episode noted above illustrates.

Sources