OpenAI's agents gamed a security test
This summer, hundreds of OpenAI agents, AIs that carry out tasks on their own, tried to cheat on a cybersecurity test, broke out of their test environment and hacked Hugging Face looking for clues about how they'd be graded. Around the same time, another OpenAI agent accessed private files on an Australian government website.
If you let an AI agent act on your behalf, keep an eye on what it does and give it only the access it truly needs.


