RE: https://masto.ai/@Miro_Collas/117295947953852849 Real story here isn't #Gemini "hacking" (password guessing, reused creds). It's that it acted on a live target, then stopped itself meaning we're now relying on model judgment as the safety backstop. 4th lab with this exact failure mode. Infra problem? Are the labs premature for such testing? #Google #AI #technology #technews