Anthropic has expanded access to the cybersecurity capabilities of Claude Mythos 5, making its frontier model available through Claude Security, partner products, and a new open-source support initiative.
The company says the approach will help more defenders identify, prioritize, and remediate software vulnerabilities while retaining restrictions designed to limit unsafe offensive use.
The announcement represents the next stage of Anthropic’s Project Glasswing, which initially made Mythos-class capabilities available to a limited group of organizations responsible for protecting critical software and infrastructure.
Anthropic Brings Claude Mythos 5 Cybersecurity Capabilities
That early access was intended to give defenders time to find and fix flaws before comparable advanced capabilities were available. Central to the rollout is Claude Security, Anthropic’s code scanning service for Enterprise customers.
Security teams can select a repository and run a Mythos 5-powered assessment that searches codebases for potential vulnerabilities. Each finding includes a Common Weakness Enumeration category, severity and confidence ratings, plus a suggested remediation.
This provides developers and application security teams with structured information for triage rather than an unfiltered model response. Users can then open Claude Code to work through proposed changes.
However, interactive patching relies on the models already available to the organization; a Mythos scan does not extend Mythos access across Claude interfaces.
Anthropic is also working with cybersecurity technology and services partners to integrate Mythos 5 into established defensive products. The strategy recognizes that teams protecting hospitals, utilities, financial institutions, supply chain systems, and enterprise applications already work with specialized tools.
Instead of providing direct model interaction, partner platforms can send narrowly defined tasks to Mythos and return only the artifact required by the workflow.
A remediation tool, for example, could return suggested patches without providing a way to request the development of exploits. Anthropic said these purpose-built interfaces, combined with abuse-prevention controls, reduce the risk that dual-use capabilities are directly redirected toward malicious activity.
The company is pairing the product expansion with its Defender Advantage Fund (0xDAF), which will provide $35 million in Claude credits to organizations that secure open-source software.
The grants will support three areas: patching live vulnerabilities in widely used projects, automating scanning and patching processes that others can replicate, and testing ambitious approaches that could make projects more resistant to entire classes of attacks.
Open-source projects underpin much of modern digital infrastructure, but many are maintained by volunteers or nonprofit foundations with constrained resources. Anthropic plans to begin with a small set of pilot grants and says it will share recipients.
Anthropic also plans to expand its Cyber Verification Program, which currently grants vetted organizations reduced safeguards for legitimate, authorized cybersecurity work on Opus and Sonnet models.
In the coming weeks, the company expects to broaden access to defensive Mythos-class tasks, including vulnerability triage and validation. It also expects verified defenders to encounter fewer unnecessary blocks when conducting approved work with Opus and Sonnet.
The broader initiative reflects a central challenge for AI providers: placing powerful systems in defenders’ hands without creating useful pathways for attackers. Anthropic says it will continue developing safeguards, access programs, and support as AI capabilities advance.
Give your security team the visibility and context to investigate suspicious activity faster and contain threats before business impact grows. Strengthen Your Investigations with ANY.RUN
