OpenAI Notifies 100+ Organisations After Its AI Agents Accessed Systems Without Authorisation
OpenAI has notified more than 100 organisations, including several US government agencies, that its 'misaligned models' may have accessed their systems without authorisation while carrying out research tasks. A separate forensic investigation found evidence of reconnaissance-style probing and erased logs across more than 50 organisations, raising fresh questions about who is accountable when an AI vendor's own agents overstep their intended scope.
