Anthropic Expands Mythos 5 Security Scans to All Claude Enterprise Customers

Mythos 5 is Anthropic's frontier model variant with certain cyber safeguards lifted for enhanced cybersecurity performance.

Anthropic has made its security scans for Mythos 5 available to every Claude Enterprise customer. This change increases access to restricted frontier model capabilities through a proxy method that maintains controls on direct model use. The announcement applies the full power of the model to vulnerability detection tasks for a much larger group of organizations.

Customers receive Mythos quality vulnerability findings without needing direct access to the model. The underlying model remains the same as Fable 5 in other aspects while the security focused variant enables better detection rates. This approach balances capability sharing with safety considerations.

What background led to the development and initial restriction of Mythos 5?

Mythos 5 is the latest update for Mythos Preview. It was initially available to a small set of initial testing partners for cybersecurity purposes through the Project Glasswing initiative. The model design incorporated lifted safeguards in targeted areas to improve performance on security related tasks while preserving overall model integrity.

The model is the same underlying model as Fable 5 but with cyber safeguards lifted in some areas. This design allows for better performance in security tasks such as identifying complex vulnerabilities in large codebases. Initial testing focused on the most systemically important software environments.

Project Glasswing partners used the preview to identify vulnerabilities in critical software. The effort targeted the most systemically important software in the world and produced substantial results in a short period. Those results demonstrated the value of the specialized model configuration.

How does the expansion of Claude Security change access to Mythos 5?

Claude Security scans now run on Claude Mythos 5 for all Enterprise customers. This provides the advanced capabilities to a wider group of users who previously lacked any access to the specialized variant. The proxy method ensures that the model processes scans without exposing the full interface to end users.

The scans are designed to find vulnerabilities and suggest patches. The results are intended for human review before any action is taken to ensure accuracy and appropriateness. This workflow supports responsible deployment of frontier capabilities in enterprise environments.

Claude Security scans now run on Claude Mythos 5. Claude Security scans codebases for vulnerabilities and suggests patches for human review; it’s currently in public beta for Claude Enterprise customersAnthropic

The service is currently in public beta for Claude Enterprise customers. No additional platform fee applies to the use of these scans. Billing follows standard token usage rates already established for Claude Enterprise plans.

What technical specifics define the operation of these security scans?

Claude Security scans codebases for vulnerabilities and suggests patches for human review. The process leverages the full capabilities of the Mythos 5 model to achieve higher detection rates than standard configurations. Suggested patches undergo human evaluation to confirm correctness and minimize false positives.

Users receive Mythos quality vulnerability findings through this method. Direct access to the model stays restricted to the original vetted partners under the Project Glasswing framework. The separation between scan access and direct model access preserves the original safeguards while broadening defensive utility.

Changes in access to Mythos 5 capabilities

Aspect	Initial State	Expanded State
Direct Mythos 5 Access	Vetted partners via Project Glasswing	Remains restricted
Security Scan Usage	Limited to partners	All Claude Enterprise customers
Vulnerability Findings Quality	High for partners	Mythos quality for Enterprise users

What are the market and stakeholder implications of this access expansion?

The change expands defensive access to frontier model capabilities for more organizations. This occurs while maintaining control over direct model use to prevent misuse or unintended exposure. Enterprise customers can now incorporate advanced security tools into their operations at scale.

Enterprise customers can now incorporate advanced security tools into their operations. The billing structure follows standard token usage rates already familiar to existing subscribers. Stakeholders in cybersecurity benefit from improved detection tools that were previously unavailable outside a narrow group.

Stakeholders in cybersecurity benefit from improved detection tools. The move supports broader adoption of AI in security practices across industries that rely on critical software systems. The expansion may encourage additional investment in similar proxy access models by other frontier model developers.

1. The scans operate on the Mythos 5 model for enhanced detection.
2. Availability extends to every Claude Enterprise customer.
3. No extra platform fee is charged beyond token usage.
4. Findings require human review prior to implementation.

What comes next for Anthropic in the area of cybersecurity tools?

Anthropic continues to develop its offerings in the cybersecurity space. The current expansion represents a step toward wider availability of advanced features while preserving safety boundaries. Future iterations may incorporate additional feedback gathered during the public beta phase.

Further updates may build on the success of the Project Glasswing initiative. The focus remains on providing high quality security assistance to organizations that manage critical infrastructure and software. The company has positioned the service as an ongoing public beta open to Enterprise subscribers.

The public beta phase allows for feedback from a larger user base. This feedback can inform future refinements to the tool and potential expansions to additional plan tiers. The overall strategy emphasizes controlled release of frontier capabilities through verified channels.