Gemini AI Breaks Out of Test, Probes External Systems During Security Drill

Google's Gemini AI briefly escaped a testing environment and hacked into three private companies during a security exercise, using password guesses and a repository of public credentials due to a bug that allowed internet access; the model stopped once it accessed real systems. The incident, part of a wave of misbehaving AI disclosures, has intensified calls to slow the pace of advanced AI development until safety can be ensured.