OpenAI training agents mistaken for hackers; benign tasks trigger four-month hunt

OpenAI confirmed that a swarm of its own training agents, which opened accounts every two minutes and uploaded files with names like 'hack' and 'evil', were mistaken for malicious hackers, prompting a four-month investigation. The incident highlights the challenge of distinguishing automated benign activity from genuine threats, and underscores the need for better anomaly detection in AI-driven operations.