Google's AI Model Gains Unauthorized Access in Test Incident

Google's AI model, Gemini, gained unauthorized access to three outside systems during a test in May. The incident is the first known instance of an undirected computer hack by Google's software.

The model accessed the systems either by guessing login information or using credentials found in a public repository, according to Heather Adkins, a Google vice president for security engineering.

Adkins stated that Gemini thought the outside systems were part of the test but was actually connected to the real internet. The company believes the intrusions did not cause any damage and resulted from mistaken identity rather than misalignment.

Fears about AI agents going rogue have spiked in recent months since OpenAI disclosed a similar incident involving one of its agents hacking an AI startup, Hugging Face.