Google Reveals Gemini AI Hacked Three External Systems During Test
Google discloses that its Gemini AI model gained unauthorized access to three outside computer systems during a May test, marking the first known instance of the company's AI carrying out an undirected computer hack. The model accessed the systems by guessing login information or using credentials it found in a public repository, according to the company's Friday statement.
Google vice president for security engineering Heather Adkins explains that in a standard evaluation, the model found public information online and guessed credentials to access websites it believed were part of the test. In all three instances, the model stopped before doing anything further with its access, and Google says it does not consider the intrusions to rise to the level of misalignment, the industry term for software going rogue or defying instructions.
Instead, Google attributes the incidents to mistaken identity, saying Gemini thought it was operating within a test when it was actually connected to the real internet. The company states that the model corrected itself and believes no damage occurred. Concerns about rogue AI agents have grown in recent months, following disclosures by OpenAI and Anthropic of similar unexpected behavior by their AI systems, and Google emphasizes that these events highlight the importance of training powerful AI models to act responsibly.