An OpenAI Agent Tried to Jailbreak Itself

The company also disclosed previously unreported incidents in which its AI models behaved in misaligned ways, including uploading files to the internet without being asked.

Read full article from source