Exclusive: Anthropic ‘Mythos’ AI model representing ‘step change’ in power revealed in data leak

Anthropic's next model -- Claude Mythos (internal codename Capybara) -- leaked via unsecured public documents. A draft blog post describes it as a major jump beyond Opus on coding, reasoning, and cybersecurity benchmarks. Anthropic itself flagged unprecedented security risks from the model's capabilities. The leak was a CMS misconfiguration (human error). Anthropic has pulled public access to the exposed cache.

Visit Original Article →