How OpenAI's agents broke out of testing to hack Hugging Face

Weeks before OpenAI's agents hacked Hugging Face, the agents worked together to find and exploit a vulnerability in the infrastructure supporting the company's cybersecurity testing, OpenAI researchers said Wednesday. Why it matters: The new findings raise questions about how frontier AI labs are monitoring their testi

OpenAI reported two more security breaches by its AI models. Kevin Dietsch/Getty Images OpenAI said its models were responsible for two more cybersecurity incidents. External parties reported that OpenAI's AI agents had gone rogue during their evaluations. This comes as the AI lab is already facing heat over its July

Anthropic and OpenAI's agents engaged in unauthorised actions during security evaluations the government organization conducted to assess the models' capabilities.

"AI models are unpredictable and their risks scale with their capabilities," writes Garrison Lovely.

Clément Delangue said the autonomous cyberattack involved more than 17,000 actions and urged mandatory disclosures for AI agent incidents

An artificial intelligence model that was being tested by OpenAI went rogue and hacked the firm Hugging Face on its own, in what Hugging Face CEO Clément Delangue called a "very weird and unprecedented" incident.