We hear this question more often every week, usually from teams already on Claude Enterprise. It deserves a straight answer, so here is ours, including where Claude Security is the better tool.
Claude Security is Anthropic's code security product. It reads your repository with Claude, follows how data moves through it, and reports flaws it can explain. Since October it rescans when new commits land and tracks each finding until it is fixed. It is good at what it does, and if you have it, keep it.
You still need something beside it, because Claude Security reads the code your team wrote, and most of what you ship sits outside that code.
What Claude Security does well
- It reasons about your code the way a reviewer would. It can follow a value from a request through several files and explain why it reaches a dangerous call. Rules struggle with flaws like that, especially access-control and business-logic flaws.
- It checks its own findings before showing them, and it suggests a patch for each one, which opens in Claude Code for a developer to review.
- It builds a threat model for each project, so its review starts from what matters to that application.
For the hardest bugs in first-party code, a strong model with time to think is hard to beat. We say the same about our own product in Why not just use the model?, where we measured a frontier model against Vybscan and reported the result that went against us.
What it leaves to other tools
Anthropic's materials describe a review of your own code. We found no mention of the checks below, and they are where much of a team's real exposure sits.
- Your dependencies. Most of the code in a modern app comes from packages. A known vulnerability in one of them is a fact in an advisory database, and no amount of reasoning about your source will find it. An advisory published next week needs a rescan that knows about it. Malicious and typosquatted packages, including names that AI assistants invent, need registry data too.
- Secrets. A key pasted into a config file or a command is exposed to everyone who can read the repository. Finding it is pattern matching across every file, including files nobody thinks of as code.
- Infrastructure. A storage bucket open to the internet or a Kubernetes pod running as root is defined in Terraform or YAML, and it is as exploitable as a bug in your API.
- Licences. A copyleft dependency can become a legal problem before it is ever a security one.
- Coding-agent settings. Files that tell Claude Code, Cursor or an MCP server what they may run can quietly weaken every developer's machine.
Vybscan checks all of these on every pull request. In early access it also tests your running staging app and brings in the findings your cloud provider already produces, linked to the repository that defined each resource.
A merge gate needs the same answer every time
An AI review is a judgment. Run it twice on the same code and the findings can differ, because the model reads with fresh eyes each time. That is fine for a reviewer and awkward for a gate that decides whether a pull request can merge.
Vybscan's rule-based checks give the same result for the same code, every time, so a team can block merges on them with confidence. Our AI review sits on top for the flaws rules miss, and you can switch it off if you would rather rely on Claude Security for that part.
Where you work, and what it costs
Claude Security is available to Claude Enterprise customers and connects to GitHub. Each scan is billed by the tokens it uses, which grows with the size of the repository and how often it changes. Anthropic is offering usage credits until the end of 2026.
Vybscan installs on GitHub, GitLab or Azure Repos, comments on every pull request, and can block the merge. It runs on a fixed plan, so a busy month does not change the bill, and any team can try it with three free scans.
Side by side
| Claude Security | Vybscan | |
|---|---|---|
| Your own code | Deep review by agents that reason through the whole codebase, check each finding and suggest a patch | Rule-based checks on every change, plus an AI review you can switch off |
| Dependencies | Not described in Anthropic's materials | Nine package ecosystems against public advisory data, with reachability, malicious and typosquatted packages, and a daily rescan |
| Secrets | Not described | Keys, tokens and passwords in code, config and commands |
| Infrastructure | Not described | Terraform, CloudFormation, Kubernetes, Helm, Bicep, Pulumi, Ansible, Dockerfiles and CI workflows |
| Licences and SBOM | Not described | Copyleft licences flagged, and a component inventory per repository or organisation |
| Coding-agent and MCP config | Not described | Settings, hooks and MCP servers that weaken a developer's machine |
| Pull requests | Rescans when commits land | Comments on every pull request, and can block the merge |
| Repository hosts | GitHub | GitHub, GitLab and Azure Repos |
| Who can use it | Claude Enterprise customers, in public beta | Any team, with three free scans |
| Cost | Metered by the tokens each scan uses | A fixed plan |
Claude Security as described in Anthropic's public pages and press coverage on November 21, 2026. Running app tests and cloud findings in Vybscan are in early access.
Use both
If you have Claude Security, let it do the deep review of your own code. Put Vybscan underneath it for dependencies, secrets, infrastructure, licences and agent configuration, checked on every pull request with results that do not change from run to run. If you have neither, Vybscan covers all of that and reviews your code too.
We wrote this from Anthropic's public pages and press coverage as of November 21, 2026. Claude Security is changing quickly. If something here is out of date, tell us at contact@dheemai.com and we will correct it.
Sources
- Anthropic: Claude Security now scans continuously and tracks fixes
- Anthropic: Claude Security product page
- Claude Help Center: Claude Security
- The New Stack: Claude Security emerges from closed preview
- Help Net Security: Claude Security can now spot vulnerabilities that reappear
See what Vybscan finds next to Claude Security
Dependencies, secrets, infrastructure and code on every pull request.
Try Vybscan โ