Anthropic has moved its most cyber-capable model into a product security teams can switch on themselves. As of August 21, 2026, Claude Security scans run on Claude Mythos 5, the Mythos-class model that until now reached only vetted defenders.

The product is not a prompt box

The idea behind the release is about interface, not capability. A scan connects to a GitHub repository, traces data flows across files and returns findings. The user cannot ask the model anything; what comes back is a fixed scan artifact rather than a chat window.

Anthropic's reasoning is that risk concentrates where a user can prompt the model directly. The capability that finds a vulnerability is the same capability that writes an exploit for it. When the model runs in the background over code you own and returns a defined output, there is no path to ask it for an exploit. The same logic extends to partner integrations: the end user receives a patch list or an alert generated by Mythos but never steerable by the person receiving it.

How the scan works

Rather than pattern-matching against rules, the model traces data flow across files and reads Git history. Anthropic says each finding then passes an adversarial verification step in which the model challenges its own result before surfacing it — the stated mechanism for cutting false positives.

Findings come back with:

  • A CWE category — the standard classification for software flaws.
  • Confidence and severity ratings — how certain the finding is and how critical.
  • A suggested fix — a patch proposal that still depends on human approval.

The target class is high-severity work: memory corruption, injection flaws, authentication bypasses and cross-file logic errors. Results can be pushed to Slack or Jira via webhooks, or exported as CSV or Markdown. Scans can be scoped to a directory and scheduled. Patching happens separately and runs on whatever models the organization already has; Mythos access does not extend to other surfaces.

Who can use it

The feature is in public beta for Claude Enterprise customers. There is no separate model add-on; Mythos 5 scans are billed as standard token usage under the existing plan. An admin enables it in the admin console. Teams on Pro, Max and Team plans do not get the Mythos-powered scan. Anthropic names hospitals, utilities, financial systems and the software supply chain as the defender constituency.

Two more things in the same announcement

The company also launched the Defender Advantage Fund, offering $35 million in Claude credits to organizations securing open-source software. It said it will expand its Cyber Verification Program over the coming weeks to cover broader dual-use capabilities on Opus and Sonnet, with Mythos-class access to follow.

The trade-off is explicit. Mythos-class models sit above the Opus tier and were withheld for a reason: Anthropic states that Mythos 5 has the strongest cybersecurity capabilities of any model in the world. This release is an attempt to hand that capability to defenders without handing attackers a new option.