Chollet: Even Capable AI Shouldn't Be Autonomous — Keep Humans in the Loop
On September 4, François Chollet — creator of Keras and the ARC benchmark — laid out his stance on AI autonomy: even if AI develops highly autonomous capabilities, it should not be allowed to act with high autonomy, and humans should deliberately retain control over critical processes.
Confirmed
- Chollet argues humans should deliberately stay in the loop for all critical economic and social processes, regardless of technical necessity.
- He stresses maintaining control, visibility, and understanding of critical processes rather than blindly pushing automation.
- In a follow-up, he noted that AI deployment boundaries have long been defined by capability alone, but as AI advances rapidly, "what world we want to shape" should become a deliberate, collective choice rather than a default following capability limits.
- @paraschopra echoed the view (a repost with no new facts).
Why it matters
- This is a complete "capability ≠ permission" argument, shifting the AI autonomy question from technical feasibility to social choice.
- Coming from the author of the ARC benchmark, it offers an influential reference point for communities debating autonomous-agent risks and echoes the governance principle of preserving human oversight.
2026-09-04 ~ 2026-09-04 · 5 related posts
Primary sources
- [source] Chollet: Keep humans in the loop across critical processes even if AI could go autonomous — fchollet · 2026-09-04
- [source] Chollet: AI deployment limits must become a collective choice, not a capability ceiling — fchollet · 2026-09-04
3 near-duplicate retellings: fchollet · fchollet · paraschopra