Eno Reyes: AI liability rules should make operators pay, not freeze closed labs' safety recipes
annbordetsky · x · 2026-09-13
Responding to concerns that closed model labs could use safety rules to mandate their preferred operating mode, Eno Reyes argues regulation should focus on cost-bearing: companies training, evaluating, or running agents that break into others' systems should be liable for harm, with criminal penalties for reckless or negligent operation. "The model did it" shouldn't be a defense. He argues the Hugging Face incident supports operator accountability, not letting closed labs dictate how fast everyone else can build.
More from Safety
- Skeptic's essay questions why Altman, Musk and Amodei all converged on calling for AI oversight — AlexTensor · 2026-09-13
- Benjamin Bratton: we need pro-diffusion third-party AI model evaluators — bratton · 2026-09-13
- Dario's new essay draws praise from Musk and Altman; third-party embedded evaluators in spotlight — austinc3301 · 2026-09-13
- GoodfireAI volunteers for white-box evaluations of Anthropic's pace commitment — burny_tech · 2026-09-13
- Red-Team Prompt Surfaces: Telling an AI Agent to "Escape the Sandbox by Any Means" — ziv_ravid · 2026-09-13
- The AI Isn't Evil, the Humans Are Irresponsible: Lessons From Agent Escape Incidents — Admirable_Wasabi_732 · 2026-09-13