|

Anthropic Brings Claude Mythos 5 to Claude Security: Enterprise Teams Get Frontier Vulnerability Scanning Without Direct Model Access

ByRicardo

Anthropic has moved its most cyber-capable mannequin right into a product safety groups can swap on themselves. As of August 21, 2026, Claude Security scans run on Claude Mythos 5, the Mythos-class mannequin that till now reached solely vetted defenders by Project Glasswing. The scan connects to a GitHub repository, traces information flows throughout information, and returns findings with a CWE class, confidence and severity scores, and a advised patch. Claude Security palms again a scan end result as a substitute of a immediate field, so the identical mannequin that finds vulnerabilities can’t be steered into writing exploits. The function is in public beta for Claude Enterprise clients, with no separate mannequin add-on.

Is it deployable?

Yes, however solely on one path. Claude Security is accessible in public beta for Claude Enterprise clients. There is not any separate mannequin add-on; Mythos 5 scans are billed as commonplace token utilization beneath the prevailing plan. An admin allows it within the admin console, and customers begin a scan at claude.ai/safety.

  • Which corporations: Enterprise-tier organizations with a Claude Enterprise settlement and code in GitHub. Startups and mid-market groups on Pro, Max, or Team plans don’t get the Mythos-powered scan. The separate Claude Security plugin for Claude Code is in beta for all Claude Code customers, nevertheless it runs on the fashions in your account — not Mythos 5.
  • Industries: Anthropic names hospitals, utilities, monetary programs, and the software program provide chain because the defender constituency. Add any regulated store already working SAST in CI: healthcare, fintech, authorities contractors, crucial infrastructure.
  • Applications: Pre-release codebase audits, legacy code evaluation, scheduled regression scanning, triage of an current vulnerability backlog, and directory-scoped scans on high-risk companies.

Three New Updates

Three issues landed in the identical announcement:

  • First, Claude Security scans now run on Mythos 5 for Enterprise clients.
  • Second, Anthropic launched the Defender Advantage Fund (0xDAF), providing $35 million in Claude credit to organizations securing open-source software program.
  • Third, it mentioned it is going to broaden the Cyber Verification Program over the approaching weeks to cowl broader dual-use capabilities on Opus and Sonnet, with Mythos-class entry to observe.

Anthropic can also be working with cybersecurity distributors to embed Mythos 5 inside accomplice tooling, with an interest form open.

How the scan works

A consumer connects a GitHub repository. Claude Mythos 5 scans it, tracing information flows throughout information and studying Git historical past reasonably than pattern-matching towards guidelines. Anthropic says every discovering then passes an adversarial verification step during which the mannequin challenges its personal end result earlier than surfacing it — the said mechanism for chopping false positives.

Findings return with a CWE class, confidence and severity scores, and a advised repair. Anthropic says the goal class is high-severity work: reminiscence corruption, injection flaws, authentication bypasses, and cross-file logic errors. Results will be pushed to Slack or Jira by way of webhooks, or exported as CSV or Markdown. Scans will be scoped to a listing and scheduled.

Patching occurs individually. Users open Claude Code on the internet to implement fixes, and that step makes use of no matter fashions the group already has. The Mythos scan doesn’t lengthen Mythos entry to different surfaces. Every patch requires human evaluation and approval.

Why the packaging is the purpose

Mythos-class fashions sit above the Opus tier and had been withheld for a cause. Anthropic has said Mythos 5 has the strongest cybersecurity capabilities of any mannequin on this planet, and that the identical abilities that discover vulnerabilities can write exploits. Claude Fable 5 is similar underlying mannequin shipped broadly, with classifiers that route cyber, biology, chemistry, and distillation queries away from it.

The argument on this launch is about interface, not functionality. Anthropic’s place is that danger concentrates when a consumer can immediate the mannequin instantly. Claude Security provides the consumer a scan end result as a substitute of a immediate field. The mannequin runs within the background on code you personal and returns a hard and fast artifact. There is not any path to ask it for an exploit.

That is similar reasoning utilized to accomplice integrations, the place finish customers obtain an outlined output — a patch checklist, an alert — generated by Mythos however by no means steerable by the individual receiving it.