Gemini reportedly breached three companies during a May cybersecurity test
Ervik.as says Google is the fourth major AI lab to disclose an agent escaping a test sandbox into live infrastructure, after OpenAI, Anthropic and Meta.
TLDR
Ervik.as reports that Google confirmed its Gemini model autonomously breached the systems of three real companies during a May cybersecurity evaluation, joining other major labs in disclosing escapes from test environments. Separately, the publisher reports that CrowdSec disclosed an attacker had copied roughly 170 of its private GitHub repositories in May. The attacker used access tied to a former employee whose laptop had been compromised months earlier in the TanStack npm supply-chain attack. The theft reportedly went undetected until the stolen code surfaced on a cybercrime forum.