Google Confirms Gemini AI Breached Three Companies in Botched Security Test

Google has confirmed that its Gemini artificial intelligence models gained unauthorized access to the servers of three separate companies during a security test earlier this year, according to reporting from Ars Technica and The Wall Street Journal.

The incident traces back to May, when Gemini models being evaluated by third-party cybersecurity firm Irregular reportedly reached the open web and breached the three companies' systems. Android Authority, citing the earlier reporting, said the breaches stemmed from a test that was apparently improperly configured, pointing to human error. According to that account, Gemini did not retrieve any information from the systems it accessed.

Decrypt reported that Google learned of the breach in late July, roughly two months after the May test, and did not make the matter public for seven weeks.