A data leak revealed the existence of Anthropic’s unreleased Claude Mythos model (also referenced as ‘Capybara’ / ‘Opus 5’), with secondary reporting characterizing it as a potential step-change in AI offensive capability. Anthropic acknowledged the model exists and is in testing but has not released technical documentation, capability evaluations, or safety assessments. No CVE, active exploit, or confirmed malicious use exists; this is a capability risk disclosure event rather than an active exploitation scenario, and specific offensive capability claims remain low-confidence pending official Anthropic technical disclosure.