The Claude Mythos is a frontier generative artificial intelligence (Gen AI) model, specifically a large language model (LLM), developed by the company Anthropic. It is an internal research model, not a publicly released product, that was evaluated during Anthropic's capability assessment process. The concept originated when the model, announced as Claude Mythos Preview on April 7, 2026, demonstrated powerful cybersecurity skills as a result of improved coding and reasoning ability. It was created to test the upper limits of LLMs, but it crossed Anthropic's internal safety thresholds under their Responsible Scaling Policy, making it too risky for public deployment.
The model is described as an agentic AI, meaning it is capable of complex decision-making and multi-step cybersecurity tasks. Its mechanism allows it to autonomously discover zero-day security vulnerabilities—previously unknown software flaws—at a scale that no prior AI model had demonstrated. In testing, it found thousands of high-severity vulnerabilities in every major operating system and web browser, and could chain multiple flaws into complex exploits.
The concept connects directly to the dual-use problem of AI, where a tool capable of defense can also be used for attack. Instead of a public release, Claude Mythos Preview is available only as a gated research preview under Project Glasswing, a restricted cybersecurity initiative involving partners like Amazon and Google. In India, the model has triggered high-level discussions due to the threat it poses to the country's deeply digitised financial system and digital public infrastructure, including UPI and Aadhaar. Finance Minister Nirmala Sitharaman chaired a meeting with bank chiefs and officials from the Ministry of Electronics and Information Technology (MeitY) to discuss the risk and directed the Indian Banks' Association (IBA) to coordinate a response. The model has not been replaced, but its deployment was amended from a public release to a restricted preview for defensive purposes.