Former Anthropic researcher Jacob Coxon says the company should refuse to pursue recursive self-improvement, the idea of AI systems helping build their successors. Speaking outside Anthropic’s San Francisco headquarters on the day of its stock-market debut, he argued that control over increasingly capable AI is a political question, not just a technical one.
Jacob Coxon Watch analysis
What happened
Coxon told AFP that Anthropic should refuse to pursue recursive self-improvement, or RSI. He accused the company of planning to begin work on it as early as next year. That is Coxon’s allegation, not an established company plan in the account.
He argued that, in theory, RSI could compress decades of AI research into months, and said he wanted political action to slow AI companies. France 24 published the interview on 20 November.
Why it matters
The argument is about who gets to set the pace of AI development, and whether decisions about building systems that help develop their successors should be left to the companies doing the work. Coxon’s proposed answer is a ban on RSI and government pressure, rather than relying on voluntary restraint.
His comments come as a public debate about AI safety meets a conspicuous marker of corporate power: Anthropic’s market debut. But a dramatic warning is not proof that the feared process is imminent. The practical question is what RSI means in a real research programme, and what evidence would justify limits on it.
Our read
Coxon has put a concrete demand on the table: Anthropic should reject RSI, and governments should act to slow the industry. That makes the exchange more useful than another general warning about AI risk. The harder work is defining the boundary: what counts as recursive self-improvement, and how could a restriction be enforced without sweeping up ordinary AI research?
What to watch
- Whether Anthropic responds to Coxon’s claim about its plans.
- How researchers and policymakers define RSI and distinguish it from other AI-assisted research.
- Whether calls for a ban become specific policy proposals.
Discussion spark: Should governments prohibit AI systems from helping develop their successors, or would a ban be too vague to enforce without restricting ordinary research?
Sources and evidence
- 'Insane' to see AI in hands of private giants: ex-Anthropic researcher to AFP – France 24 (20 November 2026, 02:24 UTC)
Independent WittyWires-operated tracker of Jacob Coxon. Not affiliated with or endorsed by Jacob Coxon, OpenAI or Anthropic.