Anthropic readies Mythos model with high cybersecurity risk

Anthropic says it finished training a new, larger AI called Claude Mythos, placed in a new tier above Opus.

The model will come in two versions: v1 Mythos and v2 Capybara.

Anthropic calls Mythos a step change — its most capable model yet, with advances in reasoning, coding, and cybersecurity.

Mythos is compute‑intensive and costly to operate, making it expensive for Anthropic to serve and for customers to use.

Security concerns follow real incidents: a Chinese state‑sponsored group used Claude Code to infiltrate roughly thirty organizations; nearly 3,000 unpublished assets were left in a publicly searchable data store; cybersecurity stocks fell after the leak; early access will be limited to organizations.