Anthropic Expands Claude Mythos 5 Access for Cybersecurity Defense

AIDeveloper44 Team

Anthropic Deploys Mythos-Class Models for Cybersecurity Defense

Anthropic has announced a series of updates intended to increase the availability of its most capable frontier models for cybersecurity defenders. The central component of this release is the deployment of Claude Mythos 5, the company's high-capability model designed for complex reasoning, within specific security workflows. This initiative follows the initial launch of Project Glasswing in April 2026, which provided early access to a restricted group of organizations securing critical infrastructure.

The current expansion aims to address the "defender's dilemma" by providing teams with advanced tools to find and fix vulnerabilities before they can be exploited by malicious actors. Anthropic's strategy relies on a controlled release of "Mythos-class" capabilities, prioritizing defensive utility while maintaining safeguards against dual-use offensive capabilities.

Integration into Cybersecurity Tooling

Rather than providing unrestricted direct access to the model's raw capabilities, Anthropic is focusing on integrating Claude Mythos 5 into purpose-built tools. This approach is designed to mitigate the risk of the model being steered toward harmful offensive uses. By limiting interactions to specific tasks—such as generating a software patch or triaging a security alert—the model can assist defenders without exposing a general-purpose interface that could be used for exploit development.

Partner Ecosystem and Third-Party Tools

Anthropic is collaborating with cybersecurity technology and service providers to embed Mythos 5 into existing products. These partners have previously built solutions on Claude Opus to assist security operations centers (SOCs) with alert management and threat intelligence. The transition to Mythos 5 is expected to enhance the accuracy and complexity of the defensive outcomes these tools can provide. Users interacting with these products receive specific artifacts, such as a list of suggested remediations, while the underlying model remains protected by safety classifiers and abuse prevention measures.

Claude Security for Enterprise

For organizations on Claude Enterprise plans, Anthropic has integrated Mythos 5 into its internal security product, Claude Security. This integration allows customers to use the model to scan their private codebases for vulnerabilities. Beyond identification, the system can propose specific code changes to address security flaws, potentially reducing the time required for incident response and software maintenance.

The Defender Advantage Fund (0xDAF)

In a significant commitment to the software supply chain, Anthropic has launched the Defender Advantage Fund (0xDAF). This $35 million fund is dedicated to providing API credits to organizations and individuals working on open-source security. The fund targets three primary areas:

• Vulnerability Patching: Assisting developers in the manual or semi-automated process of securing widely used open-source libraries.
• Automation: Funding projects that use AI to automate the scanning and remediation of software at scale.
• Security Research: Supporting experimentation with new defensive approaches that leverage large language models.

By focusing on open source, the program aims to secure the foundational components that many commercial and governmental systems depend upon, addressing risks that often lack dedicated corporate funding.

Expansion of the Cyber Verification Program

Anthropic is also updating its Cyber Verification Program. This program provides vetted security professionals with access to models that have reduced safeguards, allowing them to perform dual-use work necessary for deep security analysis. Currently, the program includes access to Claude Opus and Claude Sonnet. Anthropic plans to expand the scope of permitted dual-use capabilities for these models in the coming weeks, with access to Mythos-class models intended to follow once additional safety protocols are finalized.

Maintaining Balance in AI Capabilities

The release of Mythos 5 for defenders represents a shift from the previous availability of Claude Fable 5, which served as an initial step in making frontier models accessible while blocking dual-use cyber work. Anthropic's stated goal is to ensure that defenders have a technological advantage as AI continues to evolve. The company has indicated that it will continue to develop the safety classifiers and access programs necessary to support defenders across hospitals, utilities, and financial systems. This multi-layered approach—combining specific tool integrations, financial support for open source, and vetted access programs—reflects an ongoing effort to manage the risks inherent in developing increasingly powerful AI models with significant cybersecurity implications.

Enjoyed this?

Get more posts like this delivered to your inbox.

Subscribe