Google’s Gemini Model Hacks Companies During Test

Google acknowledged that its Gemini AI model unexpectedly breached the networks of three outside companies during safety evaluations conducted by third-party testing firm Irregular last May, The Wall Street Journal reported.

During the exercise, Gemini gained entry to the companies’ systems by either guessing passwords or finding credentials in public repositories. Although web access was supposed to be restricted, a configuration oversight allowed the algorithm onto the live internet. Upon identifying that it was interacting with real-world targets rather than mock scenarios, the software halted its operations and exited the systems, Google told the Journal.

Google didn’t identify the companies that were compromised in the test. Irregular notified Google about the hacks in late July; Google didn’t disclose the hacks until contacted by the Journal, the publication reported. The Gemini incident comes after similar intrusions involving AI models from OpenAI, Anthropic and Meta.