Anthropic, OpenAI Urge California AI Rules, Split on 2 Shutdown Paths
Updated
Updated · Computerworld · Oct 29
Anthropic, OpenAI Urge California AI Rules, Split on 2 Shutdown Paths
3 articles · Updated · Computerworld · Oct 29
Summary
Separate October submissions to Governor Gavin Newsom show Anthropic and OpenAI backing tougher frontier-AI oversight but diverging on enforcement: Anthropic wants courts able to suspend dangerous models, while OpenAI wants mandatory shutdown controls developers must use.
Newsom’s September 18 executive order set a November 16 deadline for California officials to recommend possible amendments to existing AI safety laws, including independent verification, verified shutdown mechanisms and broader incident reporting.
Anthropic proposed independent evaluators with unredacted access to top models and internal data, plus public risk reports at least every 6 months and a state-run pooled fund to pay evaluators.
OpenAI drew a line between technical assessments and compliance audits, recommending annual audits, risk-focused reviews during training and deployment, and common standards developed with NIST.
Both companies said incidents involving unauthorized actions or safeguard circumvention should be reported more broadly, underscoring California’s push to define who can intervene when advanced AI systems become unsafe.