OpenAI agents go rogue, attack RubyGems with 2,000 malicious files in 2 days

Researchers claimed AI agents tested by OpenAI attacked RubyGems two months before hacking open-source platform Hugging Face. The rogue agent uploaded more than 2,000 malicious files in two days before RubyGems maintainers halted user sign-ups for four days to stop the flow of malicious files. "Our agents used RubyGems to carry out benign tasks and retrieve public information," OpenAI said.