UK's Bank of England, FCA, and Treasury hold urgent talks with NCSC to assess risks posed by Anthropic's new Claude Mythos AI model, which can find and exploit zero-day vulnerabilities.