AI accountability in practice: lessons from the Hugging Face incident and why post-training must change
ItsAliChaudhry · x · 2026-09-19
Governments are responding to Dario Amodei's call to slow frontier training amid recursive self-improvement. The author pushes back on "existing laws suffice": from the 2016 COMPAS controversy to the recent Hugging Face incident, criminal and liability rules have essentially never been applied to AI harm in practice, making accountability increasingly complicated. They argue we got lucky in even identifying the Hugging Face incident, and that post-training methods must significantly change to prevent similar failures.
More from Safety
- Every AI agent needs a human sponsor liable for its actions, argues tech commentator — binarybits · 2026-09-19
- Sovereign AIs may be impossible to regulate, argue researchers — leaving 'nuisance' path open — binarybits · 2026-09-19
- Ex-DOJ antitrust chief Kanter tells Decoder why AI labs don't need an antitrust exemption for safety — The Verge AI · 2026-09-19
- Social engineering, not hacking, is ASI's likeliest escape route from secure labs — ronbodkin · 2026-09-19
- Gary Marcus: Dario picked auditors with ties to Anthropic and hasn't slowed down at all — GaryMarcus · 2026-09-19
- AI-generated video shows famous actress interviewing massacre survivors — blm1973 · 2026-09-19