Anthropic PBC accidentally leaked details of its new Claude Mythos large language model through a content management system error, revealing 3,000 publicly accessible assets including draft launch materials.
The company confirmed Mythos represents its most capable model to date, with engineers completing training and beginning pilot programs with early customers. The model will be priced higher than existing Claude offerings.
Mythos introduces a fourth product tier to Anthropic's lineup, surpassing the current Claude 4.6 Opus in programming and reasoning tasks. The Capybara variant specifically excels at cybersecurity vulnerability detection.
Anthropic plans early access releases to help organizations strengthen code security against AI-driven exploits. Company engineers already discovered over 500 high-severity vulnerabilities using current Claude tools.
Cybersecurity stocks including CrowdStrike Holdings and Palo Alto Networks dropped over 5% following the announcement, as investors anticipate competition from Anthropic's advanced detection capabilities.
The revelation coincides with OpenAI's completion of its new 'Spud' model pretraining, setting up intensified competition in the enterprise AI market.