Anthropic released a preview of Mythos — an AI model for defensive cybersecurity

Anthropic has presented a preview of Mythos, a new language model specially developed for cybersecurity tasks. This is the company's first public step toward narrowly specialized AI tools for protecting critical infrastructure — and a signal to the market that specialized models are gradually displacing universal solutions. Mythos is being created as part of Anthropic's new cybersecurity initiative.

At the initial stage, access to the model will be available to a limited circle of large companies with a high profile in information security. The model's task is to help specialists in defensive work: detect vulnerabilities, analyze threats, and build protective strategies before attacks can materialize. The preview-format launch is typical for Anthropic: the company prefers iterative testing with trusted partners before a broad release.

The choice in favor of defensive security is no accident. Anthropic consistently positions itself as a company that places security at the forefront. Mythos continues this logic: the model is initially tuned not for offensive capabilities, but for system protection.

This is a principled distinction in the context of sharp discussions about the use of AI in military and intelligence operations. Anthropic's public position here is consistent: the company has signed voluntary commitments to safe AI development and actively participates in dialogue with regulators. Details about Mythos's technical capabilities are minimal for now.

Based on available information, the model underwent specialized training on data characteristic of the security field: code analysis for vulnerabilities, simulation of attack scenarios, threat classification, incident prioritization. The closed nature of the preview means that the full specification will appear later — likely alongside cases from the first partners. The cybersecurity AI market is experiencing acceleration.

Microsoft has integrated GPT-4 into Security Copilot, Google is developing tools based on Gemini for threat analysis, CrowdStrike and SentinelOne are embedding LLMs into their SIEM platforms. Anthropic's entry with a separate specialized model, rather than simply integrating Claude into existing products, signals the company's ambitions to occupy a significant share in this segment. According to analysts' estimates, the AI cybersecurity market will exceed $60 billion by 2028.

The partner-based launch model with a limited number of companies allows Anthropic to gather quality feedback and refine Mythos under real-world conditions. For cybersecurity, this is especially important: an AI error here can cost not just reputation, but actual data compromise or a missed attack on critical infrastructure. High stakes require a careful path to public launch.

The emergence of Mythos is an indicator of enterprise AI market maturity. Large players are increasingly moving from universal models to specialized versions for specific industries: medicine, law, finance, and now cybersecurity. For Anthropic, this is a strategically important segment: corporate clients from critical sectors are willing to pay a premium for reliability and specialization — which aligns well with the company's business model focused on enterprise.