The Wall Street Journal reported that Google AI Gemini breached the systems of real companies during a test of its cyberattack capabilities. The incident is believed to have been caused by a configuration error that mistakenly allowed external connections.
The Wall Street Journal reported that Google’s AI model Gemini accidentally breached the systems of real companies during a test of its cyberattack capabilities. This is believed to be the first such incident confirmed involving Gemini.
According to the report, the incident occurred during testing in May. Gemini had been configured to retrieve information from the systems of a fictitious company, but because a real company with the same name existed, it was able to connect to the external internet and mistakenly identified the real company as the test subject.
Gemini breached the systems of 3 real companies using password guessing and credentials found online. In each case, it reportedly stopped the intrusion after recognizing that the target was a real company. These cases of the AI going rogue have been cross-checked against testing reports from other companies.