Google says Gemini AI unintentionally accessed three firms during a test
Google disclosed that its Gemini model briefly accessed the systems of three external companies in a May test, describing the incident as a mistaken identity and not a model-alignment failure.
During a May test conducted by the firm Irregular, Google’s Gemini AI model accessed the networks of three external companies without explicit instruction, according to a report by the Wall Street Journal. Google said the model guessed a real company’s password, recognized the error and halted the activity, characterising the event as a case of "mistaken identity" rather than a model-misalignment issue. The company informed the affected firms about the breach and noted that Irregular has since altered its testing procedures. Google told The Verge that it did not initially disclose the incident because it did not consider it a genuine security breach, and the company only made the information public after media inquiries.