OpenAI expands TAC program, challenges Anthropic’s Mythos model

OpenAI is expanding its Trusted Access for Cyber (TAC) program in response to Anthropic’s Mythos model, and the probability of Anthropic having the third best AI model by April 30, 2026, now sits at 21.5% YES.

OpenAI’s decision to broaden access contrasts with Anthropic’s selective release of Claude Mythos Preview, which is limited to about 40 organizations. OpenAI’s wider distribution could weaken Anthropic’s competitive position, particularly if Mythos continues to be viewed as too risky for general availability. The Anthropic third best AI model market reflects this, with odds dropping as OpenAI pursues its broader access strategy.

The TAC expansion includes the GPT-5.4-Cyber model, optimized for defensive tasks, now reaching thousands of verified individuals and teams. This open-access approach works directly against Anthropic’s exclusive distribution model. With 14 days left until resolution, the April 30 odds on Anthropic’s third-place ranking are increasingly bearish.

Trading volume for this market is currently light, with no face value reported, suggesting traders are cautious. But small volumes can still produce sharp swings in thin markets, and any new developments could move odds quickly.

Anthropic restricting Mythos to a select group leaves room for OpenAI’s broader strategy to capture more of the cybersecurity market. At 21.5¢, a YES share pays $1 if Anthropic secures the third spot by April 30, a 4.65x return. For that to happen, Mythos would need to outperform competitors on key benchmarks within a tight window.

Watch for announcements from benchmark evaluators at LMSYS Chatbot Arena or new performance leaks from Google DeepMind or OpenAI. Either could move these odds fast.

Get prediction market intelligence as a structured API feed. Early access waitlist.