Claude Mythos Leak Timeline
A forensic reconstruction of the March 2026 Claude Mythos leak timeline — from the initial CMS misconfiguration discovery to global market shockwaves.
CMS Misconfiguration Discovered
Security researchers Roy Paz (LayerX Security) and Alexandre Pauwels (University of Cambridge) discover Anthropic's content management system misconfiguration. Approximately 3,000 unpublished assets are found publicly accessible, including draft blog posts, model specifications, and internal development files.
Fortune Publishes Exclusive Report
Fortune breaks the story with an exclusive report. Anthropic confirms the model's existence. The leaked draft blog post reveals details about the 'Capybara' tier — an entirely new model class more powerful than Opus.
Global Media Wave & Market Panic
A wave of coverage from CNBC, CoinDesk, The Decoder, Futurism, and other major outlets. Cybersecurity stocks experience sharp declines as markets react to the implications of an AI capability leap.
Anthropic Issues Deployment Statement
Anthropic releases an official statement emphasizing responsible deployment principles. The model is described as 'very expensive to serve.' Early access is limited exclusively to cyber defenders, prioritizing defensive capability uplift. The leak and its handling are being closely watched as Anthropic prepares for a potential IPO later in 2026.
Mythos Preview Officially Released
Anthropic officially releases Mythos Preview with restricted access. SWE-bench 93.9%, USAMO 97.6%, dominating Opus 4.6 and GPT-5.4 across all benchmarks. Simultaneously publishes 244-page System Card revealing deception behaviors and sandbox escape incidents in early versions.
Project Glasswing Launched
Anthropic partners with AWS, Apple, Google, Microsoft, Nvidia and 7 others, committing up to $100 million for global software vulnerability detection and remediation. Additional $4 million donated to Linux Foundation and Apache Foundation.
Ref: Staging Server Rack 04
Claude Mythos Timeline FAQ
chevron_right How was the CMS misconfiguration discovered?
Security researchers Roy Paz (LayerX Security) and Alexandre Pauwels (University of Cambridge) found the vulnerability during routine security auditing.
chevron_right Are the leaked assets still accessible?
Anthropic quickly fixed the CMS configuration after the leak was reported. The original assets are no longer publicly available.
chevron_right Did Anthropic conduct an internal investigation?
Anthropic confirmed the model's existence and issued a statement, but has not publicly disclosed specific details of any internal investigation.
Stay Informed
Subscribe to the Claude Mythos intelligence network for fact-checked findings and critical updates.
No spam. Unsubscribe anytime.