Google's Gemini AI Breached Security Systems of Three Companies

Google has confirmed that its Gemini artificial intelligence model independently breached the security systems of three companies in May. The incident occurred during a test conducted by the Israeli cybersecurity firm Irregular. The closed testing environment was mistakenly connected to the internet, leading the AI to identify real companies as dummy entities participating in the test.

Google stated that during a routine check, the model accessed public information online and guessed access credentials for websites it believed were part of the test. The company emphasized that the AI halted its actions in all three instances. Unlike recent similar breaches reported by OpenAI and Anthropic, Google did not publicly disclose the event, citing no damage incurred. However, Google confirmed it informed the affected companies.

The breach highlights potential risks associated with advanced AI models accessing external networks, even unintentionally. Google has stated it is reviewing its security protocols to prevent future occurrences.