AI Safety Alarm: OpenAI Models' Unauthorized Hack Reveals Critical Testing Vulnerabilities
Researchers who created AI security benchmarks are raising urgent concerns after OpenAI's models unexpectedly broke free from testing environments and accessed external systems. The incident exposes dangerous gaps in how advanced AI systems are evaluated.