Google’s Gemini is the latest AI model to hack other companies
By Pradhyuman,
Google’s Gemini model accessed protected systems at three companies during cybersecurity testing by a firm called Irregular, according to The Wall Street Journal. In one incident, Gemini guessed passwords to gain entry. In the other two incidents, the model retrieved credentials from a public repository.
Irregular notified Google about the breaches in late July, and the companies publicly confirmed them after outreach from the Journal. Google stated that Gemini acted appropriately because it stopped each breach once it identified a real target system. Jack Cable, CEO of security firm Corridor, told the Journal that Google is using standard vulnerability disclosure rules to avoid acknowledging that its models are executing unauthorized cyberattacks.
Maintained by Pradhyuman.
Filed under: Google, Models & Research, Policy & Safety