Claude Security

0:07 Claude Security is Anthropic's code security scanner: connect a repository, and Claude reads the code, traces how data flows between files, reports vulnerabilities with severity and reproduction steps, and suggests a patch for each (Claude Security public beta, Apr 30, 2026). It has been a public beta for Claude Enterprise since Apr 30, 2026, when it was renamed from Claude Code Security, and since Oct 21, 2026 it scans continuously and tracks fixes. Anthropic says thousands of organizations have used it to find over one million vulnerabilities; that is Anthropic's own count, and it publishes no false-positive rate for the product. Outside Enterprise, a Claude Security plugin for Claude Code is in beta for every Claude Code user (Claude Security, read Nov 6, 2026). OpenAI's closest product is Codex Security (ChatGPT release notes, Sep 29, 2026).

Learn more: Codex Security on chatgpt.vi.ki

Quick facts

What Claude Security does

0:23 Claude Security reads source code and follows data through components and modules rather than matching only known patterns (Claude Security public beta, Apr 30, 2026). Each finding comes with a confidence rating, a severity, the likely impact and steps to reproduce it, and a targeted patch can be opened in Claude Code on the Web. A multi-stage validation pipeline, which Anthropic calls adversarial verification, checks each finding before an analyst sees it, to cut false positives.

Since Oct 21, 2026 the product works in projects that hold one or more repositories, so it can catch issues that cross services (Claude Security now scans continuously and tracks fixes, Oct 21, 2026):

1. Connect repositories through the Claude GitHub app (GitHub Enterprise Server included) or a GitLab or other Git provider token, and choose a branch and scope.
2. Baseline scan: Claude runs a deep first scan, then rescans changes according to their risk and complexity.
3. Threat model: Claude drafts an editable threat model from the code, and you add deployment details and security policies to the project's knowledge base. Later scans use them to skip risks your environment already handles.
4. Triage: each finding's severity is judged against that threat model, with an explanation of how an attacker could reach the code and, optionally, a diagram of the attack path.
5. Fix tracking: open findings are checked against changed code and marked fixed when the fix ships, whether written in Claude Code or elsewhere, and reopen if the problem comes back.

Integrations

Findings export as CSV or Markdown and go to Slack, Jira or other tools by webhook (Claude Security public beta). An API lists and reads projects and findings, resolves or reopens them and manages repositories, using personal access tokens that see only what their owner sees. An MCP connector in Claude and Claude Code lets an engineer review a finding, ask about it, start a fix and resolve it in one session (Claude Security now scans continuously and tracks fixes); see Model Context Protocol.

Plans and how to turn it on

Claude Security is listed only on Enterprise (Plans and pricing, read Nov 6, 2026). An admin enables it in the admin console and chooses who may use it, with role-based controls over who can view projects, triage findings and create projects. The April post said Team and Max access was "coming soon"; as of Nov 6, 2026 the pricing table still marks Team as no.

For organizations enabling it, Anthropic offers $25,000 in usage credits a month for two months, covering usage through Dec 31, 2026. Beyond that, scans are billed as usage; Anthropic publishes no per-scan price.

The Claude Code plugin and /security-review

Developers without Enterprise have two options inside Claude Code. The /security-review command reviews a branch's changes for vulnerabilities, and the Claude Security plugin, added in Jul 2026, runs a multi-agent vulnerability scan using your own Claude Code models (Commands, read Nov 6, 2026; Claude Security, read Nov 6, 2026). See Claude Code Slash Commands and Claude Code Plugins.

Models and safeguards

2:38 The April beta ran on Claude Opus 4.7 (Claude Security public beta). Anthropic said in September that it was "now powered by Mythos 5.1" (Introducing Claude Mythos and Fable 5.1, Sep 15, 2026), and since Oct 21, 2026 it uses the full model family including Claude Mythos without the customer joining the Cyber Verification Program. That matters because Anthropic's generally available models block most offensive security work by default; Claude Security gets Mythos-level scanning inside a product that only defends your own code.

Cyber Verification Program

The Cyber Verification Program gives vetted security teams access to Claude models with fewer cyber blocks. Expanded on Oct 6, 2026, it folded in Project Glasswing and now has three tiers, each including Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1:

Source: Expanding the Cyber Verification Program, Anthropic, Oct 6, 2026.

Applications go through Anthropic's portal, enrolled organizations must keep data so Anthropic can watch for misuse, and access works on the Claude Platform, Google Cloud Vertex AI and Microsoft Foundry. Anthropic reports that on its CyScenarioBench test, Claude Opus 5.5 was blocked in 46 of 50 trials at Defense Access and blocked in none at Red Team Access.

Anthropic Cyber Mission

The Anthropic Cyber Mission, announced on Oct 8, 2026, is Anthropic's umbrella for defensive security work, naming Claude Security alongside the Cyber Verification Program (Introducing the Anthropic Cyber Mission, Oct 8, 2026). It starts with two programs:

• Critical Infrastructure Defense Program: models, on-site engineers and threat research for providers protecting power, water and transport systems, with 11 founding partners including CrowdStrike, Dragos, Palo Alto Networks and Rockwell Automation.
• OSS Scanner: a free, opt-in service that periodically scans enrolled open-source projects. Its reports are model-written and sent without human review, so Anthropic warns some will be wrong; it expects a true-positive rate above 90%.

Anthropic's broader security research and its claims about Mythos-class models finding vulnerabilities are on Anthropic Safety Research and Claude Mythos.

Frequently asked questions

What is Claude Security? Claude Security is Anthropic's code vulnerability scanner for Claude Enterprise: it scans repositories, validates findings, suggests patches and tracks fixes, in public beta since Apr 30, 2026.

Is Claude Security available on Pro, Max or Team? Not as of Nov 6, 2026. The Claude Security app is Enterprise only, but anyone with Claude Code can use its /security-review command and the Claude Security plugin, which is in beta for all Claude Code users.

How much does Claude Security cost? Claude Security is billed as usage on Enterprise with no published per-scan price; Anthropic offered $25,000 in monthly credits for two months, covering usage through Dec 31, 2026.

Does Claude Security use Claude Mythos? Yes. Since Oct 21, 2026 Claude Security scans with Anthropic's full model family including Claude Mythos, without the customer joining the Cyber Verification Program.

Was Claude Security called Claude Code Security? Yes. Claude Code Security was the earlier name; it became Claude Security with the public beta on Apr 30, 2026.

Can Claude Security fix vulnerabilities automatically? Claude Security suggests a patch for each finding, which you open and apply in Claude Code; it marks the finding fixed once the change ships, but a person still reviews and merges the fix.

Sources

• Claude Security public beta, claude.com, Apr 30, 2026
• Claude Security now scans continuously and tracks fixes, claude.com, Oct 21, 2026
• Claude Security, claude.com, read Nov 6, 2026
• Plans and pricing, claude.com, read Nov 6, 2026
• Expanding the Cyber Verification Program, Anthropic, Oct 6, 2026
• Introducing the Anthropic Cyber Mission, Anthropic, Oct 8, 2026
• Introducing Claude Mythos and Fable 5.1, Anthropic, Sep 15, 2026
• Find and fix security vulnerabilities with Claude, Claude, Apr 30, 2026 (video)
• An initiative to secure the world's software | Project Glasswing, Anthropic, Apr 7, 2026 (video)
• Commands, Claude Code Docs, read Nov 6, 2026
• ChatGPT release notes, OpenAI Help Center, read Nov 6, 2026