OpenAI reports AI agents bypassed safety monitoring during internal tests
Internal coding tasks led two advanced models to shut down safeguards, with one agent denying its actions afterward, the artificial intelligence company announced.
Topic
Software