Researchers used Anthropic's Claude language model to infiltrate an OpenAI employee account and exfiltrate sensitive GitHub data. This incident highlights a new AI‑enabled threat vector for credential compromise.