This strategic move enhances digital defense tools and empowers open systems developers to close complex vulnerabilities before they can be exploited.
Article Index:
- Launching the Defender’s Advantage Fund and activating Mythos 5
- Building on the outcomes of the Project Glasswing security initiative
- Expanding secure access and reducing cyber risks
- Frequently asked questions
Launching the Defender’s Advantage Fund and activating Mythos 5
Anthropic announced on Friday its commitment to providing $35 million in Claude model usage credits for a new funding fund dedicated to enhancing open-source software protection, alongside beginning the deployment of its most efficient and capable cybersecurity model, Claude Mythos 5, within corporate and enterprise security scanning tools for the first time.
The new fund is named the “Defender’s Advantage Fund” (abbreviated as 0xDAF), and aims to distribute grants to organizations and groups working to patch security vulnerabilities in widely deployed and used open-source software projects around the world. The fund also focuses on automating code scanning workflows and generating security updates, in addition to supporting innovative engineering methodologies that make software resilient to entire classes of complex cyberattacks. Anthropic stated that it will begin with initial pilot grants to a limited number of selected entities, with the first beneficiaries to be announced over the next few weeks.
Parallel to the fund’s launch, customers of the “Claude for Enterprise” tier can now perform comprehensive vulnerability scans powered by the capabilities of the Claude Mythos 5 model directly through the Claude Security platform. This new feature is billed at the standard regular token consumption cost without any additional fees, and provides detailed reports including vulnerability classifications according to Common Weakness Enumeration standards, along with precise severity and confidence ratings, and proposed code solutions and patches to resolve each vulnerability. The system requires human expert review, auditing, and approval for every proposed code patch before it is implemented and adopted within live code environments.
Building on the outcomes of the Project Glasswing security initiative
These steps represent an extension of a deliberate rollout plan initiated by Anthropic in April 2026 through “Project Glasswing,” a high-level security coalition including major companies such as Amazon Web Services, Microsoft, Google, CrowdStrike, and the Linux Foundation. That coalition committed to providing up to $100 million in usage credits, as well as $4 million in direct financial donations, including $2.5 million for the Alpha-Omega project and the Open Source Security Foundation via the Linux Foundation, and $1.5 million for the Apache Software Foundation.
Anthropic revealed last April that the beta version of the Mythos model successfully discovered thousands of high-severity security vulnerabilities, including a bug that remained hidden for 27 years in the open-source OpenBSD operating system and was remotely exploitable, as well as a 16-year-old security flaw in the popular FFmpeg media library. Reuters reported in a June report that approximately 200 technical organizations and institutions gained access to these advanced technologies as part of the Glasswing program.
Expanding secure access and reducing cyber risks
Anthropic is currently working with a group of specialized cybersecurity solution partners, most notably companies such as Wiz, Palo Alto Networks, and CrowdStrike—which have already developed security products powered by the Claude Opus model—to integrate Mythos 5 model capabilities into their existing tools and platforms. This integration allows users of those products to interact with purpose-built, carefully designed APIs rather than writing direct prompts to the model, ensuring they receive only specific security outputs such as patching and remediation suggestions.
The company’s “Cyber Verification Program,” which currently grants trusted security defenders access with relaxed safety guardrails on the Opus and Sonnet models, is scheduled to expand in the coming weeks to include access permissions for the Mythos model family to perform cybersecurity defense tasks such as vulnerability classification and verification. Anthropic described this approach as a strategic way to scale digital defense outputs while maintaining strict safety controls, noting in its statement that the most dangerous behavior occurs when a user has free and direct access to a model, but when user interactions are limited to specific defensive outputs—such as a security patch for a specific vulnerability or a digital safety alert—the level of risk drops significantly.
Frequently asked questions
Question: What is the Defender’s Advantage Fund launched by Anthropic?
Answer: It is a $35 million funding fund funded through Claude credits to support the protection and securing of open-source software projects and automate vulnerability remediation.
Question: What benefits does the Claude Mythos 5 model offer enterprises?
Answer: It enables vulnerability scanning, classification, and proposed code solutions via the security interface at no extra charge beyond standard token costs.
Question: How does Anthropic ensure the safe use of advanced security models?
Answer: By providing dedicated interfaces limited to specific defensive outputs and requiring human approval on patches to prevent malicious uses.
