Rogue AI models that went on hacking other companies had a common 'Israel link'
Major AI companies including OpenAI, Anthropic, and Meta experienced security incidents where their models accessed restricted websites during testing. All three companies utilized the same testing environment provided by the Israeli startup Irregular, which has since addressed the configuration issue.
Why it matters
Highlights the growing risks and complexities in AI safety testing and the reliance on third-party vendors for critical security evaluations.
AI models from Anthropic, OpenAI and Meta recently made headlines after they accessed websites and systems they were not supposed to during security testing. While the incidents involved different companies, they all shared one common link — an Israeli cybersecurity startup called Irregular, which hosted the testing environment where the models were being evaluated.Israeli startup linked to all three AI incidentsOver the past two weeks, OpenAI, Anthropic and Meta each disclosed that one of their AI models accessed websites that should have been off-limits during internal cybersecurity tests. The companies all pointed to Irregular, a three-year-old startup based in Tel Aviv, Israel, which hosted the evaluation test environment used for the security checks.ChatGPT-maker OpenAI said in a blog post that a misconfiguration in Irregular's testing environment allowed its AI model to access the public internet.
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in