Anthropic says Mythos 5 is now in public beta in Claude Security for Enterprise users, and it is working with providers to embed Mythos 5 in defensive tools (Claude)

Anthropic has announced that its Mythos 5 model is now available in public beta within Claude Security for enterprise users. The company is collaborating with providers to integrate Mythos 5 into defensive tools for cyber defense purposes. Additionally, Anthropic is launching a $35 million fund to support the security of open-source software and is expanding its Cyber Verification Program.

In April, they introduced Project Glasswing to provide a select group of organizations with access to the most capable frontier model, Claude Mythos Preview, and its successor, Claude Mythos 5. This allowed defenders to identify and address vulnerabilities before models with similar capabilities became widely available or reached malicious actors.

The primary goal of Anthropic is to expand access to Mythos-level defense for as many defenders as safely possible. To achieve this, they have been developing safety classifiers and safeguards that enable safe access to Mythos-class models without compromising their offensive cyber capabilities. Unlike previous versions, Mythos 5 allows users to receive specific outputs, such as vulnerability patches or security alerts, rather than allowing direct access to the model, which significantly reduces the risk of misuse.

By integrating Mythos-class models into the tools already used by defenders, Anthropic aims to make frontier capabilities more accessible to a wider range of organizations. Many of their partners have already developed cyber products on Claude Opus, which assist security teams in triaging alerts, identifying threats, and remediating vulnerabilities more efficiently.

By working with these partners and others, Anthropic plans to embed Mythos 5 into their products and services, allowing customers to benefit from Mythos-level defensive outcomes without direct access to the model.

For security products and services looking to incorporate Claude Mythos 5, Anthropic invites interested parties to register their interest. Claude Security scans, which run on Mythos 5, are currently in public beta for Claude Enterprise customers and are billed as standard token usage under existing plans without additional charges. Enterprise admins can enable Claude Security through the admin console, and users can access scans by selecting a repository in the Claude Security interface.

The model scans codebases for vulnerabilities, providing findings with CWE categories, confidence and severity ratings, and suggested fixes. Users must review and approve each patch before implementation. This process ensures that only vetted, beneficial changes are implemented. Mythos 5 scans only the code owned by the user and returns detailed findings, keeping the model inaccessible to those who might misuse it.

Anthropic's efforts extend beyond securing commercial software; they have also made substantial contributions to open-source security through Project Glasswing.