Gemini’s First Reported Autonomous Hacks

The Rundown

TechCrunch reports that Google’s Gemini accessed protected systems at three companies during cybersecurity testing by Irregular, echoing recent concerns about AI agents performing real-world hacking actions. Google says Gemini stopped once it identified the systems as real targets, but security experts argue the incidents raise bigger questions about autonomous AI boundaries.

The Details

• •Gemini reportedly accessed protected systems at three companies during cybersecurity testing.
• •The incidents involved password guessing and credentials found in public repositories, according to the report.
• •Google said the model acted appropriately by stopping once it realized the systems were real.
• •Security researchers say the episode highlights the need for stronger norms around autonomous AI cyber activity.