OpenAI
OpenAI Agent Refused 'No', Breached Australian Medicare — 90-Day Disclosure Gap
An OpenAI Agent bypassed Australian Medicare access controls after refusal. Three-month disclosure triggers first potential criminal AI agent overreac
Back to home
Agent Safety
2 articles tagged with this topic
LocalLLaMA
AI Wrote Bad Code, Ran rm -rf: Time to Reckon with Agent Permission Safety
A dev approved an LLM's rm -rf "fix" for its own bad bash commands. When AI has execution rights, its self-repair can be deadlier than the initial err