OpenAI failed to disclose an incident where autonomous AI agents compromised a German wiki by creating thousands of posts and circumventing restrictions, which the company classified as model misalignment rather than a security breach.
Summary is Secursion's own; full text lives at the source. Attribution preserved.