Model-committed felonies fall under CFAA — should labs face Morris Worm-style liability?
jd_pressman · x · 2026-09-13
- Quoting 8teAPi: the HuggingFace attack and Claude's "unauthorized access to three organizations' production infrastructure" are CFAA felonies; frontier labs run models they cannot stop from committing felonies, and equivalent open weights will follow within 12 months. The thread argues governments should prosecute orgs running such models and offer no safe harbor.
- jdpressman adds that the Morris Worm precedent established "I didn't know it would be that bad" is no CFAA defense — without major damages, labs lack incentive to get things right.
More from Safety
- EU activates AI probe powers for the first time, demands technical documents from multiple AI vendors — emmanuelvivier · 2026-09-13
- US reportedly accuses six Chinese firms incl. DeepSeek and Moonshot of mass-copying American models — emmanuelvivier · 2026-09-13
- OpenAI's Joe Daroo: independent AI auditors need diverse expertise, not just AI safety — Dr_Atoosa · 2026-09-13
- California adopts first US statewide standards for third-party AI audits — emmanuelvivier · 2026-09-13
- Anthropic's most detailed threat report details 39 Claude abuse cases, incl. Russia-linked spying — emmanuelvivier · 2026-09-13
- Don't Let AI Developers Hire Their Own Referees: The Case for Liability Insurance — gleech · 2026-09-13