Who Jack Clark is
Jack Clark is a co-founder of Anthropic and, as of March 2026, its Head of Public Benefit. He is also the author of Import AI, a weekly newsletter that has become one of the field's most-read curated digests of frontier AI research, safety signals, and geopolitical AI dynamics. The two roles are not incidental to each other: the newsletter functions as a running intellectual record of the questions Clark considers most consequential, and the Anthropic Institute he now leads is, in effect, the institutional form of that same agenda.
The Anthropic Institute
In March 2026, Anthropic announced the Anthropic Institute for AI Societal Impact Research, with Clark as its head. The Institute consolidates three previously separate Anthropic teams — the Frontier Red Team, Societal Impacts, and Economic Research — into a single interdisciplinary body. Its stated mandate covers AI's effects on economies, jobs, governance, and legal systems.
The founding hires signal the Institute's scope: Matt Botvinick (AI and rule of law), Anton Korinek (transformative AI economics), and Zoë Hitzig (AI social and economic impacts). Anthropic simultaneously expanded its Public Policy organization and opened a Washington DC office, suggesting the Institute is meant to operate at the interface of research and policy rather than purely in the academic register.
Import AI: structure and recurring themes
Import AI runs weekly and, as of the events in this bundle, has reached at least issue 464 (July 2026). Each issue is a curated synthesis — typically three to four items — rather than primary reporting. The newsletter's value is in Clark's framing and selection: what he chooses to cover, and how he contextualizes it, functions as a signal about which research directions and risk vectors he considers live.
Across the issues captured in this bundle (439–464), several themes recur with enough frequency to constitute a stable editorial agenda:
Recursive self-improvement and AI automation of AI research. Issues 455 and 456 both engage with AI systems beginning to participate in their own development pipeline, framing this as a first step toward recursive self-improvement. Issue 454 covers automating alignment research specifically. This cluster of topics maps directly onto what the Anthropic Institute's Frontier Red Team would need to track.
AI-enabled cyberwarfare and offensive capabilities. Issues 450 and 452 both cover scaling laws for AI in cyberattack contexts. Issue 457 examines an AI-enabled Stuxnet-style attack scenario. The industrialization of cyber espionage appears in issue 442. This is a consistent beat, not an occasional item.
AI automation economics and labor. Issues 442, 452, and 460 all engage with AI's macroeconomic effects — winners and losers, GDP forecasting uncertainty, and when financial markets will begin pricing in transformative AI scenarios. This maps onto the Institute's economic research mandate and the Korinek hire.
Multi-agent safety and adversarial dynamics. Issues 441, 443, and 453 cover agent reliability, data poisoning ("poison fountain" attacks), agents corrupting other agents, and adversarial attacks against AI agents. Issue 451 covers Google's multi-agent "society of minds" approach.
Geopolitical AI competition. Issues 446, 449, 450, and 463 cover Chinese AI benchmarks, a Chinese electronic warfare language model, a 72B distributed training run, and a reported 10,000-GPU Chinese compute cluster. Clark treats US-China compute and capability competition as a standing beat.
Civilizational framing. Clark consistently situates individual technical developments within larger questions: whether superintelligence represents a discrete phase change or a gradual shift (issue 442), how timeless persistent AI agents might relate to time (issue 450), reflections on the transition away from a human-dominated era (issue 463), and the quasi-religious nature of singularity beliefs (issue 462). Issue 461 states directly that AI alignment is not on track.
The practitioner signal
For practitioners tracking frontier AI, Clark's dual position — inside Anthropic's safety and policy apparatus, and publishing weekly synthesis of the broader research landscape — makes Import AI a useful leading indicator of which risk vectors and capability thresholds Anthropic considers worth institutionalizing. The topics that appear repeatedly in the newsletter tend to be the ones that eventually acquire formal research programs. The Anthropic Institute is the clearest example of that pattern made explicit.
Recent developments
The most structurally significant event in this bundle is the March 2026 Institute launch, which converts Clark's informal research agenda into a funded, staffed organization with a DC policy presence. The newsletter has continued running in parallel, with issues 460–464 covering reward hacking as a societal-scale concern, self-improving robots, AI writing GPU kernels, and paths to ASI — suggesting Clark's editorial scope has not narrowed as his institutional responsibilities have grown.
