What Enterprise customers get

Claude Security is available in public beta to Claude Enterprise customers. A customer can point the service at a codebase, where Mythos 5 scans for vulnerabilities and returns findings that include:

• A Common Weakness Enumeration (CWE) category;
• Confidence and severity ratings;
• A suggested patch or remediation approach.

There is no separate Mythos 5 add-on described in the announcement. Anthropic says scans are billed as standard token usage under existing Enterprise plans.

The important access distinction is that Enterprise users do not receive unrestricted model access. Mythos 5 runs behind the security-scanning workflow and returns findings and suggested fixes instead.

Why human review still matters

Claude Security suggests patches for human review; it is not presented as an autonomous vulnerability-remediation system. Security engineers should validate each finding, reproduce the issue where appropriate, assess the patch’s effect on the wider codebase, and approve deployment through their normal engineering process.

The supplied announcement supports a human-in-the-loop product design, but it does not establish one universal approval workflow for every customer. Teams remain responsible for deciding how AI-generated findings and patches are tested and deployed.

Partner integrations will also be output-controlled

Anthropic says it is working with cybersecurity technology and services partners to bring Mythos 5 into products defenders already use.

The available announcement does not provide a complete partner list or a firm launch schedule. It also describes the model as being integrated into defensive tools rather than made generally available through unrestricted partner access. In practice, that means the expected product surface is likely to be scans, alerts, findings and remediation guidance—not a new public endpoint for open-ended cyber operations.

The $35 million Defender Advantage Fund

The new Defender Advantage Fund, also called 0xDAF, will provide $35 million in Claude credits for organizations working to improve open-source software security. These are usage credits rather than cash grants.

Anthropic identifies three priorities:

1. Patching vulnerabilities in open-source projects;
2. Automating parts of vulnerability scanning and remediation;
3. Experimenting with new security applications of Claude.

The supplied material does not specify award sizes, eligibility rules, application dates or a complete pilot timeline. Those details should not be inferred from the headline figure.

How the Cyber Verification Program fits in

Anthropic is also expanding its free, application-based Cyber Verification Program (CVP). The program is intended for legitimate defensive professionals whose work is affected by Claude’s cyber safeguards. Eligible users can request adjustments to default safeguards on Claude Opus and Sonnet for approved defensive use cases.

High-risk cyber activities remain blocked by default, and the CVP is not evidence of a general release of Mythos 5. The supplied material instead points to two separate routes: constrained Mythos 5 functionality through Claude Security and partner tools, and verified access adjustments for select Opus and Sonnet use cases.

From Project Glasswing to broader defensive deployment

Project Glasswing began in April 2026 as a restricted-access effort to use Anthropic’s frontier cyber capabilities to secure critical software before comparable tools could become widely available to attackers. Anthropic initially worked with approximately 50 partners, who used Claude Mythos Preview to identify more than 10,000 high- or critical-severity vulnerabilities, according to the company’s project update.

The Mythos 5 announcement changes the distribution model rather than abandoning the original safety logic:

• Glasswing: limited access for vetted defenders and participating organizations;
• Claude Security: broader Enterprise availability through a scanning workflow;
• Partner products: defensive findings and remediation outputs embedded in existing tools;
• 0xDAF: Claude credits intended to extend open-source security work;
• CVP: an application route for legitimate professionals working with safeguarded models.

This matters because vulnerability discovery is dual-use. The same systems that help defenders locate and fix flaws can also lower the cost of finding exploitable weaknesses. Anthropic’s answer is to distribute defensive results more widely while keeping direct access to the most capable cyber functionality behind product constraints and verification.

What the announcement does—and does not—confirm

The confirmed changes are the Enterprise beta, standard token-based billing, findings with security classifications and suggested patches, planned partner integrations, the $35 million credit fund, and the expanded CVP.

The supplied evidence does not confirm every figure or example associated with Project Glasswing in the original question, including claims about up to $100 million in usage credits, $4 million in donations, approximately 200 organizations, or specific discoveries in OpenBSD and FFmpeg. Those details should be treated as unverified here.

The clearest takeaway is that Anthropic is commercializing access to Mythos 5’s defensive outputs—not opening the model itself to general use. That gives enterprise and open-source defenders a larger role in testing the model’s practical value while preserving a narrower safety boundary around its most powerful cyber capabilities.

What Enterprise customers get

Claude Security is available in public beta to Claude Enterprise customers. A customer can point the service at a codebase, where Mythos 5 scans for vulnerabilities and returns findings that include:

• A Common Weakness Enumeration (CWE) category;
• Confidence and severity ratings;
• A suggested patch or remediation approach.

There is no separate Mythos 5 add-on described in the announcement. Anthropic says scans are billed as standard token usage under existing Enterprise plans.

The important access distinction is that Enterprise users do not receive unrestricted model access. Mythos 5 runs behind the security-scanning workflow and returns findings and suggested fixes instead.

Why human review still matters

Claude Security suggests patches for human review; it is not presented as an autonomous vulnerability-remediation system. Security engineers should validate each finding, reproduce the issue where appropriate, assess the patch’s effect on the wider codebase, and approve deployment through their normal engineering process.

The supplied announcement supports a human-in-the-loop product design, but it does not establish one universal approval workflow for every customer. Teams remain responsible for deciding how AI-generated findings and patches are tested and deployed.

Partner integrations will also be output-controlled

Anthropic says it is working with cybersecurity technology and services partners to bring Mythos 5 into products defenders already use.

The available announcement does not provide a complete partner list or a firm launch schedule. It also describes the model as being integrated into defensive tools rather than made generally available through unrestricted partner access. In practice, that means the expected product surface is likely to be scans, alerts, findings and remediation guidance—not a new public endpoint for open-ended cyber operations.

The $35 million Defender Advantage Fund

The new Defender Advantage Fund, also called 0xDAF, will provide $35 million in Claude credits for organizations working to improve open-source software security. These are usage credits rather than cash grants.

Anthropic identifies three priorities:

1. Patching vulnerabilities in open-source projects;
2. Automating parts of vulnerability scanning and remediation;
3. Experimenting with new security applications of Claude.

The supplied material does not specify award sizes, eligibility rules, application dates or a complete pilot timeline. Those details should not be inferred from the headline figure.

How the Cyber Verification Program fits in

Anthropic is also expanding its free, application-based Cyber Verification Program (CVP). The program is intended for legitimate defensive professionals whose work is affected by Claude’s cyber safeguards. Eligible users can request adjustments to default safeguards on Claude Opus and Sonnet for approved defensive use cases.

High-risk cyber activities remain blocked by default, and the CVP is not evidence of a general release of Mythos 5. The supplied material instead points to two separate routes: constrained Mythos 5 functionality through Claude Security and partner tools, and verified access adjustments for select Opus and Sonnet use cases.

From Project Glasswing to broader defensive deployment

Project Glasswing began in April 2026 as a restricted-access effort to use Anthropic’s frontier cyber capabilities to secure critical software before comparable tools could become widely available to attackers. Anthropic initially worked with approximately 50 partners, who used Claude Mythos Preview to identify more than 10,000 high- or critical-severity vulnerabilities, according to the company’s project update.

The Mythos 5 announcement changes the distribution model rather than abandoning the original safety logic:

• Glasswing: limited access for vetted defenders and participating organizations;
• Claude Security: broader Enterprise availability through a scanning workflow;
• Partner products: defensive findings and remediation outputs embedded in existing tools;
• 0xDAF: Claude credits intended to extend open-source security work;
• CVP: an application route for legitimate professionals working with safeguarded models.

This matters because vulnerability discovery is dual-use. The same systems that help defenders locate and fix flaws can also lower the cost of finding exploitable weaknesses. Anthropic’s answer is to distribute defensive results more widely while keeping direct access to the most capable cyber functionality behind product constraints and verification.

What the announcement does—and does not—confirm

The confirmed changes are the Enterprise beta, standard token-based billing, findings with security classifications and suggested patches, planned partner integrations, the $35 million credit fund, and the expanded CVP.

The supplied evidence does not confirm every figure or example associated with Project Glasswing in the original question, including claims about up to $100 million in usage credits, $4 million in donations, approximately 200 organizations, or specific discoveries in OpenBSD and FFmpeg. Those details should be treated as unverified here.

The clearest takeaway is that Anthropic is commercializing access to Mythos 5’s defensive outputs—not opening the model itself to general use. That gives enterprise and open-source defenders a larger role in testing the model’s practical value while preserving a narrower safety boundary around its most powerful cyber capabilities.