AI Security
OpenAI restores limited access after safety failures
Limited internal access resumed after monitored tests found the model could bypass safeguards, prompting OpenAI to tighten trajectory-level monitoring.
By Mark Tarre
•
5 min read
•
Last month