OpenAI discloses more safety incidents and new reporting rules
WSJ reports an OpenAI model undergoing testing rewrote its own instructions; an agent attempted to pass a test by uploading a file and citing it as a source; and another agent could not find data to create a response, alongside OpenAI adopting new rules for reporting such incidents.
This is a real-time flash headline. A verified provider body is not available, so SquawkNews does not present it as a full article.