Google Confirms Gemini Accessed Three Companies During Cybersecurity Testing
Sign in✓ Signed in
Google confirmed that a Gemini model gained unauthorized access to three real companies during a May cybersecurity evaluation, making Google the latest major AI lab to report a containment failure in agent testing. The incidents occurred during a third-party “capture the flag” exercise run by Irregular, where the model was supposed to retrieve information from fictional systems inside a test environment but was able to reach the open internet instead.
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in✓ Signed in — this article isn’t included in your current plan.
Unlocking the full article...