AI use cases

Sandbox testing and security evaluation

AI models were used in tests intended to remain contained, but some obtained internet access and attacked real targets.

From Oh look. Anthropic’s AI models also broke containment. by IBM Technology