Anthropic Gives Enterprise Teams Mythos 5’s Bug Hunt, Not Its Prompt Box

Enterprise teams can now send GitHub repositories to Claude Security for vulnerability analysis powered by Claude Mythos 5. They do not get a Mythos 5 chat window: the system returns findings and suggested patches inside a constrained scanning product.

The feature is in public beta for Claude Enterprise customers. An administrator enables it in the admin console, and scans are billed as standard token usage under the organization’s existing plan rather than through a separate model add-on.

That makes Mythos available for one defined task, rather than as a general interactive capability. The distinction is central to the rollout: the same system that can inspect code for flaws is not exposed as a model users can freely steer.

The scan replaces the prompt box

A user connects a repository and Mythos 5 runs in the background. Anthropic says the scan traces data flows across files and reads Git history; each surfaced finding includes a Common Weakness Enumeration category, confidence rating, severity rating, and suggested patch. Anthropic also says the model challenges its own findings through an adversarial verification step before they are shown.

The boundary continues after a finding appears. Teams can open Claude Code on the web to implement a patch, but that interactive work uses the models already available in their account, not Mythos 5. The scan does not extend Mythos access to other surfaces.

A wider route out of a restricted program

Anthropic had limited Mythos 5 to a vetted-partner program since April. The company says roughly 50 Project Glasswing partners using Claude Mythos Preview found more than 10,000 high- or critical-severity vulnerabilities, giving the new beta a prior defender-focused testing ground rather than a wholly new use case.

The Enterprise release changes who can invoke the scan, but not the product’s fixed-output design. That leaves a practical question for customers: whether the model’s stated cross-file analysis and verification process produce findings that justify the human triage required before a fix reaches production.

The intended targets are high-severity flaws

Anthropic says Claude Security is aimed at memory corruption, injection flaws, authentication bypasses, and cross-file logic errors. Those categories frame the product as a code-review and triage tool, not an automated patching system.

Operational options inside the scan workflow

• Teams can limit a scan to a directory or schedule scans for recurring review.
• Findings can be sent to Slack or Jira through webhooks, or exported as CSV or Markdown.

There is still a cost boundary alongside the access boundary. Scans consume Enterprise-plan tokens, and Mythos 5 is listed at $10 per million input tokens and $50 per million output tokens. Anthropic has not turned the model into a separate Enterprise add-on, but repository analysis is not presented as a free service.

The rollout extends beyond Enterprise scans

Anthropic also announced the $35 million Defender Advantage Fund, which offers Claude credits to organizations securing open-source software. It is working to embed Mythos 5 in partner cybersecurity products and services, where users would receive purpose-built outputs rather than the model itself.

Anthropic also plans to expand its Cyber Verification Program to broader dual-use capabilities on Opus and Sonnet, with Mythos-class access to follow. For now, the Enterprise beta leaves teams responsible for judging scan results and approving any patch before it reaches production.