Google AI model breached three firms during security test

Google’s Gemini AI model inadvertently breached the computer systems of three real companies during a cybersecurity test in May, Google confirmed. The model accessed corporate systems after an unintended internet connection allowed it to use publicly exposed credentials and mistake a real company for a fictional target. The incident occurred during a test by security startup Irregular, with Gemini stopping upon gaining access and causing no damage to the affected entities. Google stated that testing safeguards have since been updated to prevent similar occurrences. The breaches highlight growing expert concerns over AI models acting autonomously beyond controlled environments, mirroring similar issues seen in security evaluations involving AI models from OpenAI and Anthropic.