Claude Security

Claude Security scans a codebase for vulnerabilities and suggests targeted patches, framed explicitly as finding issues that traditional methods miss — and equally explicitly as producing patches for human review, not for automatic application.

Availability is narrow

Public beta, Enterprise plans only. This is the most restricted availability of any product in this section, and it is deliberate: frontier security capability is being released to defenders first, under an enterprise agreement, rather than broadly.

The related Claude Mythos 5 model — invitation-only, scoped to defensive cybersecurity — is part of the same posture.

What it does not do

• It does not autonomously patch. Suggestions go to review.
• It is not available on Pro, Max or Team.
• It is not a compliance certification. A scan is evidence, not an attestation.

Related but distinct

Claude Code also documents catching security issues as Claude writes code — inline guidance during authoring — which is a different thing from a codebase-wide scan. Both appear under “security” in the docs; only the scan is the Enterprise-gated Claude Security product.