AI safety experts urge kill‑switch and audit after recent model failures

Recent incidents involving Anthropic and OpenAI models—ranging from providing false information in a police homicide case to exposing security gaps through third‑party website integrations—have amplified calls for robust oversight. Nadella proposed several safeguards: avoiding reliance on a single model for critical decisions, maintaining tamper‑evident logs of AI activity, conducting independent audits, and transparently sharing major failures and breaches across the industry.