What is Claude Security?
Claude Security scans your code for security flaws the way a reviewer would rather than the way a rules engine does. It reads the repository, writes a threat model you can edit, follows data across files and services, and checks each finding several times before showing it, with a severity rating and an explanation of how an attacker would reach the code. Clicking Fix hands the finding to Claude Code.
It began as Claude Code Security in February 2026, entered public beta for Claude Enterprise on 30 April 2026, and on 21 October 2026 gained continuous scanning of new commits, fix tracking, an API, webhooks and an MCP connector. Scans run on Claude Mythos, Anthropic's cybersecurity-focused model, which users cannot call directly. Anthropic says it has found more than a million vulnerabilities across thousands of organisations.
It is still Enterprise-only. Pro, Max, Team and API users can run a separate Claude Security plugin inside Claude Code, which uses whatever models the account has.
How much does Claude Security cost?
| Plan | Price | What you get |
|---|---|---|
| Claude Enterprise | $20/seat/month | Plus usage at token cost; premium seat required |
| Launch credits | Up to $25,000 a month | Two months, through 31 December 2026 |
Claude Enterprise only ($20/seat/month plus usage), with a premium seat per user. Scans bill at token cost with no platform fee; eligible organisations get up to $25,000 a month in credits for two months, through 31 December 2026.
Is Claude Security worth it?
Most code scanners are good at the obvious mistakes and bad at the ones that get companies breached: the permission check that works in one service and is skipped in the next.
Claude Security goes after those. It reads the whole repository, writes a threat model of what matters in your system, and follows how data moves between files and up to eight repositories in one project. Each finding is checked several times before you see it, comes with a severity based on how exploitable it is in your code, and explains the path an attacker would take. Vanta told Anthropic it caught logic flaws none of its other tools had, and IonQ said it found material issues a penetration test run at the same time missed.
The October update made it a running service rather than a one-off audit. After a baseline scan, which can take minutes or hours, it watches new commits and rescans by risk, marks findings fixed when the fix lands, and reopens them if the bug comes back. Findings flow to Slack or Jira through webhooks, and an API and MCP connector let Claude Code pick them up.
The catch: you have to be a Claude Enterprise customer, with a premium seat for every user and usage billing switched on. Team and Max were promised access "soon" in April and still do not have it. The help centre also says scans are "stochastic by design", so two runs on the same code can return different findings, which is awkward for an audit trail. There is no list of supported languages.
Data handling rules out some buyers. Zero data retention is not supported, and Anthropic can keep data where the law requires it or to investigate misuse. You may only scan code you own outright, not third-party or outside open-source projects.
The price is the token bill and nothing else, with a separate spend limit and a per-project daily cost view. Anthropic gives no estimate per repository, though eligible organisations get up to $25,000 a month in credits for two months, through 31 December 2026, which is enough for a serious trial.
Our take: Claude Security is a second reviewer, not a replacement for the scanner in your build pipeline. Teams already paying for Claude Enterprise should run the free credits against their oldest, most tangled service and see what comes back.
Where it earns its place
An application security lead at a 400-person fintech on Claude Enterprise
Find authorisation gaps that span the API gateway and three backend services
Put the four repositories in one project, add architecture docs to the knowledge base and review the generated threat model before the baseline scan. The launch credits of up to $25,000 a month cover the trial through 31 December 2026.
A platform engineer at a 1,000-developer software company
Route new findings to the owning team without another dashboard
Webhooks send new findings to Jira or Slack, Fix opens the finding in Claude Code, and the API resolves it when the pull request lands. Fixed findings reopen by themselves if the bug returns.
A compliance officer at a regulated health insurer
Decide whether source code can go to the scanner
Claude Security does not support zero data retention, so a ZDR requirement rules it out. Anthropic points those teams to the Claude Code plugin, which can run with other model providers.
Before you commit
- Setup
- An Owner enables it under Organization settings > Claude Security; users need premium seats and usage credits switched on. Connect repositories through the Claude GitHub App, GitHub Enterprise Server or a token for GitLab, Bitbucket and other Git servers, then run a baseline scan.
- Integrations
- GitHub, GitHub Enterprise Server, GitLab, Bitbucket and other Git servers; an API with personal access tokens; webhooks for Slack or Jira; an MCP connector for Claude and Claude Code; email alerts when scans finish.
- Data and privacy
- Zero data retention is not supported, and Anthropic may keep data where the law requires it or to deal with usage policy violations. Only code you or your company own may be scanned.
- Cost at scale
- Token cost only, with no platform fee; a separate spend limit and per-project daily and month-to-date cost. No published estimate per repository or scan.
- Limits
- Up to eight repositories per project, counting context-only repositories; scans are stochastic and can return different results; large monorepos need folder scoping; GitHub Enterprise projects from the old version cannot be migrated.
What’s good
- Follows data across files and up to eight repositories to find multi-step bugs
- Editable threat model and knowledge base shape what it flags
- Continuous scans with automatic fix tracking and reopening
- API, webhooks and an MCP connector for Claude Code
- No platform fee, plus up to $25,000 a month in launch credits
What’s not
- Claude Enterprise only, with a premium seat per user
- Results can differ between runs on the same code
- No zero data retention option
- No published list of supported languages
- No published cost per scan or per repository
Best for
Security teams at Claude Enterprise companies who want a reviewer for logic and cross-service flaws on code they own.
Claude Security alternatives
Claude Security: common questions
- Is Claude Security worth it?
- Claude Security finds the logic bugs your scanner misses and tracks them to the fix, but only Enterprise customers can buy it and the results can change from one scan to the next. We rate it 3.9 out of 5.
- How much does Claude Security cost?
- Claude Enterprise only ($20/seat/month plus usage), with a premium seat per user. Scans bill at token cost with no platform fee; eligible organisations get up to $25,000 a month in credits for two months, through 31 December 2026.
- What are the downsides of Claude Security?
- Claude Enterprise only, with a premium seat per user; results can differ between runs on the same code; no zero data retention option; no published list of supported languages; no published cost per scan or per repository.
- Who is Claude Security best for?
- Security teams at Claude Enterprise companies who want a reviewer for logic and cross-service flaws on code they own.
- How much does Claude Security cost for a team or in production?
- Token cost only, with no platform fee; a separate spend limit and per-project daily and month-to-date cost. No published estimate per repository or scan.
- Is Claude Security safe for company data?
- Zero data retention is not supported, and Anthropic may keep data where the law requires it or to deal with usage policy violations. Only code you or your company own may be scanned.
- What are the best Claude Security alternatives?
- The closest alternatives we have reviewed are CodeRabbit, OpenAI Codex, Claude Code and GitHub Copilot.
Answers checked Nov 6, 2026.