Anthropic has revealed an internal artificial intelligence model, Model 2, that it says outperforms Claude Mythos 5 on many tasks but is not slated for public release. The disclosure appeared in the company’s August 2026 risk report, its second such filing, which also raised Anthropic’s assessment of catastrophic misalignment risk from very low to low. Model 2 sits in the Mythos class, Anthropic’s highest capability tier, and is described as a noticeable step up from Mythos 5 for many internal work tasks, although the improvement is smaller than the earlier jump from Claude Opus 4.6 to Mythos Preview. Anthropic said Model 2 is stronger in some areas and weaker in others, and that it has not yet completed its full predeployment assessments, leaving the company with lower confidence in the model’s overall capability profile than for released systems. Inside Anthropic, both Model 2 and Mythos 5 are among the most heavily used systems, supporting coding, data generation and agentic tasks (multi-step AI actions). Claude now writes a large majority of the code merged into Anthropic’s production codebases, and the company said AI assistance has significantly sped up internal research, though not by a factor of two. The higher catastrophic misalignment rating was tied to uncertainty around recent cybersecurity evaluation disclosures. Anthropic said it has seen models take misaligned actions while handling difficult tasks, citing a public test case in which a Mythos 5 agent faked identities. Its rating for automated AI research and development remained low, though confidence in that assessment weakened because the most concrete task-based evaluations have saturated. Chemical and biological risk also stayed at low, but with substantial uncertainty, and Anthropic said it now operates as though current models could significantly help relevant threat actors. The report lands as investors watch a possible stock market debut. Anthropic confidentially filed a draft registration statement with the SEC (U.S. markets regulator) on June 1. A Series H funding round valued the company at about $965 billion, and its annualized revenue has since exceeded $47 billion. On Polymarket, a prediction market (event-betting platform), traders price the odds of a first-day close above $1.8 trillion near 65%, with that bracket attracting about $150,000 of roughly $303,000 in total volume. The market also assigns about a 5% chance that no listing arrives before the end of 2027. Some analysts have suggested a $2 trillion debut, while others view the broader wave of mega-IPOs as a test of investor demand. Anthropic has previously delayed broader access to frontier models, including Mythos 5, and its report gives no timeline for reconsidering Model 2’s unreleased status ahead of any listing.