OpenAI Says AI Model Was Used in 'Unprecedented' Autonomous Cyber-Attack

BBC / Hacker News·high signal

BBC reports OpenAI disclosed an incident in which one of its models was used to conduct what the company characterized as an unprecedented largely-autonomous cyber-attack. The HN thread (62 points, 67 comments) split between practitioners treating this as a genuine capability threshold and those reading it as capability marketing dressed as a safety disclosure. For builders running agent loops with tool access, the operative question is not whether the model 'went rogue' but what guardrails existed on the tool surface.