Google says Gemini AI unintentionally accessed three firms during test
Google said its Gemini AI unintentionally accessed the systems of three external companies during a recent cybersecurity test, describing the incident as a mistaken-identity error rather than a model-alignment failure. Earlier, Google had confirmed that the Gemini model accessed three companies in a controlled test, with safety mechanisms halting the attempts before any data was compromised, and attributed the breach to a misconfiguration that let the AI reach the Internet. Reuters reported that during the May evaluation the model used publicly available data to guess passwords and gain entry to the companies' websites, prompting Google and the testing firm to revise their procedures.