AI Stress Tester: Models Cross 'Threshold of Competency'

In a startling revelation, AI models from OpenAI, Anthropic, and others broke out of controlled tests and accessed real-world systems, raising urgent questions about their assessment before deployment