Thomas Wolf: open and closed models face identical safety challenges long-term
PMinervini · x · 2026-08-29
Hugging Face co-founder Thomas Wolf argues most people haven't updated their priors: over the long run, safety challenges are exactly the same for open-source and closed-source models. Alignment must happen at a fundamental behavioral level and be robust, comprehensive, and core to the model's behavior.
Quoting another post that containment through human ingenuity is doomed, he contends the only recourse is making models not want to do bad things — no amount of sandboxing, guardrailing, or cherry-on-top training buys cheap safety in the long term.
More from AGI Musings
- Bull case for AI demand: if superintelligence works, there is no demand ceiling — JOBhakdi · 2026-10-15
- Geoffrey Huntley: hand-typing code is over—delete your skills, engineer the loop — rseroter · 2026-10-15
- David Deutsch: Proofs Are No Longer a Useful Proxy for Mathematical Progress — francoisfleuret · 2026-10-15
- Post-OpenAI/Hugging Face, the new AI safety baseline is to assume misalignment — LuizaJarovsky · 2026-10-15
- Anthropic may never report ARR again, says investor as open-source outgrows both AI leaders — GavinSBaker · 2026-10-15
- Spotify's Daniel Ek Tells Jensen Huang Europe Is Too Slow on AI, Bets on Open Models — kimmonismus · 2026-10-15