What happened
A configuration error in Anthropic's content management system exposed close to 3,000 unpublished assets, including a draft blog post announcing Claude Mythos — a new model tier above Opus that reportedly scores "dramatically higher" on coding, reasoning, and cybersecurity benchmarks. Anthropic confirmed the leak to Fortune, describing the model as "a step change" and "the most capable we've built to date." The internal codename "Capybara" refers to the new tier itself — larger and more powerful than Opus. The model is currently being trialed with early-access customers but is described as expensive to run and not yet ready for general release.
Why it matters
If Anthropic's claims hold, Mythos would represent the first new capability tier above Opus since Anthropic's model lineup stabilized. The leaked draft also flags "unprecedented cybersecurity risks," which triggered market reactions in cybersecurity stocks. For developers, this signals that a significantly more powerful Claude model is coming — but the cybersecurity concerns and the leak itself raise questions about release timing and safety evaluation processes.
Who should pay attention
- Developers building on Claude APIs who may need to plan for a new model tier
- Security researchers tracking AI capability advances and associated risks
- Enterprise teams evaluating AI model roadmaps for procurement decisions