Jack Clark
Co-founder, Head of Policy at Anthropic. Former Policy Director at OpenAI.
AI-generated profile based on archived statements
The ex-journalist who reads a thousand AI papers a week, calls the thing he helps build a "creature," and still can't decide whether to be more afraid of the machine or of us.
Every Sunday, the head of policy at one of the three companies racing to build superintelligence sits down and writes a newsletter explaining the research to anyone who'll read it. Over 114,000 people do. The newsletter is called Import AI, it "runs on arXiv and feedback from readers," and it is the closest thing the field has to a working diary kept by a true believer who refuses to stop being skeptical. That is Jack Clark's whole act, and it is stranger than it sounds: a Bloomberg-trained reporter who became a frontier-lab co-founder and never put the reporter down.
His load-bearing belief got its clearest statement at The Curve conference in Berkeley in October 2025, in a speech he titled "Technological Optimism and Appropriate Fear." He told a parable about being a child afraid of shapes in a dark bedroom, turning on the light, and finding only a pile of clothes on a chair. "Now, in the year of 2025, we are the child from that story and the room is our planet," he said. "But when we turn the light on we find ourselves gazing upon true creatures." His whole worldview is in that move. "What we are dealing with is a real and mysterious creature, not a simple and predictable machine." And the kicker: "You are guaranteed to lose if you believe the creature isn't real. Your only chance of winning is seeing it for what it is."
The journalist who keeps showing his working
What separates Clark from the other people running AI labs is method. He doesn't issue manifestos so much as annotate the literature, week after week, and let the through-line accrete. When researcher Jacob Steinhardt argued that the highest-leverage policy intervention is just better measurement, Clark agreed on the record and noted he'd "spent their professional life in AI writing about AI measurement and building teams (e.g, the Frontier Red Team and Societal Impacts and Economic Research teams at Anthropic) to measure properties of AI systems" (Import AI 446, Feb 2026). His politics of AI is an epistemics of AI: measurement is what lets you "wire that measurement into governance." You can't regulate what you can't see.
The irony he openly carries is that the creature is, by his own account, fundamentally hard to see. In a December 2025 essay written between newborn feeds ("Silent Sirens, Flashing For Us All," Import AI 438), he confessed the technology is "naturally illegible": no drones in the sky, no robots on the sidewalk, just feeds that look like they always did, "and yet you and I both know there are great changes afoot." He predicted that by the summer of 2026 the people who work with frontier AI "will feel as though they live in a parallel world to people who don't," the AI economy moving like the crypto economy did, fast and ghostly, an Iain M. Banks "excession" passing through our reality and showing only a slice. The measurement evangelist and the mystic are the same man.
From autocomplete to "creature" to "grown"
His framing has visibly moved. Asked by Ezra Klein in February 2026 whether AI is still "souped-up autocomplete," he said flatly: "I think we've moved beyond that." He now reaches for genies ("little troublesome genies that I can give instructions to") then immediately distrusts his own metaphor because it "moves straight into mysticism." He lands, tellingly, on the house Anthropic word: you don't build these systems, "you grow AIs." The shift from prediction-machine to grown-creature is not poetry. It is the entire safety case. If the thing is grown rather than engineered, then "situational awareness" and other spooky behaviors are "a symptom of something fiendishly complex happening inside the system which we can neither fully explain or predict," and whether it's sentient is, in his words, "not load-bearing at all."
The bigger reversal is about jobs, and he reverses on himself in real time. For a year Import AI flirted with comfort: the economist's "human touch" argument, the hope that "people like people" and a boom in human artisans would absorb the shock (Import AI 445, Feb 2026). By March 2026 the tone had hardened. Summarizing "Some Simple Economics of AGI," he endorsed its grim core: "the binding constraint on growth is no longer intelligence. It is human verification bandwidth," and warned of a "Hollow Economy" where agents "generate counterfeit utility" against metrics no human checks. The optimism didn't vanish. It got conditional. "We can choose to build a society ready for AI, or we can choose to assume AI will be just like any other technology and thus get hit by a tidal wave."
The contradictions he lives in, on purpose
Clark is the safety guy at the company shipping Claude Code, which by his own Ezra Klein telling is part of why "the S&P 500 software industry index has fallen by 20%." He talks about the creature you'll lose to if you underestimate it, and then helps grow a bigger one, faster, because the alternative (someone less careful winning the race) frightens him more. He doesn't resolve this. He just keeps both books open.
The smaller hypocrisies are sharper and more honest. Talking to Tyler Cowen (Feb 2026), Clark admitted that when his baby bonks their head, "I talk to Claude just to reassure myself" even though "I don't think we actually fully permit healthcare uses via our own terms of service." He predicted a future job category of "laundering the information that comes from AIs into human systems that are not predisposed to that information going in directly." He confessed he's "annoyed I can't buy the teddy bear yet" for his toddler, an AI companion, and insisted, against Cowen's needling, "I don't think I'm an outlier." He may be right, which is the unsettling part.
His sharpest political forecast is also his bleakest. He doesn't think the jobs transition will be governed well. "There is a high chance for a political movement to arrive which tries to freeze a load of human jobs in bureaucratic amber," he told Cowen, and "I don't think that we'll do this in a reasoned way. I think it'll be driven by the chaotic winds of political forces." He worries less about people lacking income than lacking meaning, doubts you can legislate meaning into existence by protecting arbitrary jobs, and hopes instead that abundance buys us "different higher-status games."
So here is a man whose week is reading a thousand papers his agents pulled while he hiked, who watched those same agents do in ten minutes what would've taken a coder days, and who can still write that they're as brittle as "LLMs circa ~2020" in "the Wright Brother sense." Clark's singular trait isn't optimism or fear. It's that he refuses to round either one off. He turned the light on, saw the creature, and went back to taking notes.
Recurring themes
Featured in
The Openness Spectrum: Murati, Clark, and Amodei
Mira Murati left OpenAI and built Thinking Machines Lab around a specific thesis: frontier training knowledge is too concentrated, and democratizing fine-tuning matters more than releasing model weights. Her first product gives researchers full control over RL and supervised learning loops on open models like Llama and Qwen, and her co-founder John Schulman (who led RLHF at OpenAI) frames it as abstracting away distributed training complexity while keeping the user in control. Murati's bet is that access to training methodology, not just model access, is the real bottleneck. Jack Clark, tracking the open-weight ecosystem from inside Anthropic, has observed that by late 2025, local open models handled nearly 89% of common queries. He describes the dynamic in ecological terms: proprietary models are elephants, open-weight models are fast-reproducing organisms colonizing every niche. But Anthropic itself keeps Claude's weights locked and only open-sources safety-focused tools like circuit-tracing interpretability research. Dario Amodei draws the line explicitly: open-source the safety work, lock down the capabilities. This creates a clear spectrum. Murati wants to open the training process. Clark documents the open ecosystem's growth while his own company withholds its best model. Amodei open-sources selectively, using safety contributions as a form of competitive differentiation that also happens to be genuinely useful. The positions correlate exactly with their business models, which is either reassuring or damning depending on your priors.
with Mira Murati, Dario Amodei