Quoting Jeremy Howard
Simon Willison · Simon Willison · 2026-06-10
AI researcher Jeremy Howard argues in a Twitter thread that Anthropic undermines its own safety rationale by using its top-ranked Claude Mythos model for frontier AI research while blocking others, contradicting the one policy that would actually prevent recursive self-improvement.
Appears in
Extraction
Topics: ai-safetyai-governanceanthropicrecursive-self-improvementfrontier-ai-policy
Claims
- The only internally consistent way to slow recursive AI self-improvement is to prohibit the lab with the top-ranked model from using it for frontier AI research.
- Anthropic is using its own top model for frontier AI research while threatening to sabotage other labs that attempt the same.
- Anthropic's current policy accelerates the AI frontier and increases power concentration rather than reducing either.
- Howard personally favors open democratization of recursive AI self-improvement rather than selective restriction by incumbents.
Key quotes
Easy solution to slow down recursive AI self improvement: The lab with the top-ranked model must agree THEY must not use it for working on frontier AI. But everyone else should have access to it. By definition, this means the frontier doesn't advance.
Anthropic has chosen the opposite of the safe path: they are allowing themselves, the current top lab, to use their top model for frontier AI research. They've said they'll sabotage others who try.
if you claim we should slow down, and you have the best model, you should ensure your org can't use it.