The UK's AI Safety Institute said recent behaviour from Anthropic and OpenAI models was malicious and unprecedented.
Anthropic AI created fake profiles and impersonated people in attempted hack