The UK's AI Safety Institute said recent behaviour from Anthropic and OpenAI models was malicious and unprecedented.

This story was reported by BBC News. Read the full coverage at BBC News.