Anthropic Prepares to Release Powerful AI Model Mythos with Enhanced Cybersecurity Capabilities

Anthropic has reported significant progress in developing its artificial intelligence model Mythos, which will become available to all clients in the coming weeks. Currently, access to Mythos is limited to participants in Project Glasswing—a consortium of partners, including cybersecurity experts and major technology companies. This restriction is due to the need to implement additional protective measures, as the model possesses advanced vulnerability detection capabilities, requiring infrastructure preparation to minimize risks.

Experts note that Mythos can detect vulnerabilities significantly faster than human specialists, though it does not surpass them in depth of analysis. For example, the model helped identify over 200 vulnerabilities in the latest Firefox browser update. However, its high operational cost—30 times higher than the previous Opus model—may limit its use by malicious actors. According to experts, such expenses are beyond the reach of most cybercriminals but could be feasible for state actors.

Simultaneously, Anthropic introduced an updated version of its Claude Opus 4.8 model, which showed improvements in tests compared to the previous version. The new model makes fewer unfounded claims and more accurately expresses uncertainty in responses. Among the innovations is the ability to adjust the model's "effort" level when generating responses: a higher level improves quality but depletes usage limits more quickly.