Claude Security now scans code continuously and marks findings fixed when fixes ship

What changed in Claude Security

Anthropic has turned Claude Security from a scanner people run on demand into one that keeps watching a codebase. In an update published on 21 October 2026, the company says the tool now "keeps up with your repositories as your code changes, keeps track of which findings get fixed, and uses what your team tells it about your environment to decide what matters."

The release bundles several changes: continuous monitoring after a baseline scan, automatic tracking of when findings are fixed, a per-project knowledge base and editable threat model, severity ratings with explained attack paths, a public API with webhooks, an MCP connector for Claude.ai and Claude Code, and role-based access controls. It is available in public beta for Claude Enterprise customers.

Anthropic also gives a scale figure for the product so far: since Claude Security entered public beta in April, it says, thousands of organisations have used it to find over a million vulnerabilities. That number is the company's own and is not broken down by severity or by how many findings were confirmed and fixed.

Continuous scanning instead of one-off scans

The core argument is that a scan captures one moment while code changes every day. Teams told Anthropic they wanted coverage that stays current without anyone having to schedule it.

The workflow starts with a project. Users connect repositories through the Claude GitHub app, including GitHub Enterprise Server, or with an access token from GitLab and other Git providers, and choose the branch and scope for each. Claude then runs a deep baseline scan and afterwards scans changes regularly, with frequency based on their risk and complexity, so the list of findings stays close to the current state of the code.

A project can hold multiple repositories. Anthropic says this lets Claude catch issues that sit between services, giving the example of a backend that trusts input a client was supposed to validate. That kind of cross-service flaw is hard for tools that look at one repository at a time.

Fix tracking and threat models

Most fixes happen outside the tool that found the issue, Anthropic notes. Claude Security now checks open findings against the code that changed and marks them fixed when the fix lands, whether the team fixed it in Claude Code or through a separate process. If the issue comes back later, the finding reopens automatically. For security teams that track remediation, that turns the scanner's list into something closer to a live backlog rather than a pile of stale reports.

Each project also gets a knowledge base where teams can describe how code is deployed, which security policies apply, and anything a new security engineer would need to know. Claude Security also generates a threat model from the code, which the team can review and edit. Later scans use that threat model to decide what is worth flagging, and the added context lets the tool skip risks the environment already handles.

Every finding comes with a severity rating, judged against the project's threat model, and an explanation of how an attacker could reach the vulnerable code, so teams can fix what is reachable first. Findings can include a diagram of the attack path. The product page adds that every finding passes adversarial verification before it reaches the team and includes a suggested patch that a human must review and approve.

API, webhooks, MCP and access controls

The update opens Claude Security to the rest of a team's tooling. A new API lets users list and read projects and findings, resolve or reopen findings, and create and manage projects and their repositories. Users create a personal access token in Claude Security settings, and that token can only see and change what the user can. Webhooks can notify other systems when a new finding lands.

Claude Security is also available as an MCP connector in Claude.ai and Claude Code. Anthropic's example is an engineer who pulls up a finding, asks Claude about it, starts the fix and marks it resolved in the same session. That pairs naturally with Claude Code Review, which became generally available last week, covering new pull requests while Claude Security covers what is already in the codebase.

New role-based access controls let admins decide who can view projects, triage findings and create projects. Anthropic says this lets organisations open Claude Security to more of the teams that own code without giving everyone permission to create projects.

Availability, Mythos access and free credits

The new version is in public beta for Claude Enterprise customers starting 21 October. Admins enable it in the admin console and set who on their team can access it. Anthropic is offering organisations $25,000 in free usage credits each month for two months, for usage through 31 December 2026, with full terms in its Help Center.

Anthropic says Claude Security is powered by its full family of models, including Claude Mythos, "with no Cyber Verification Program access required." That matters because Mythos access has otherwise been gated through the tiers of Anthropic's Cyber Verification Program. The product page clarifies the limit: Mythos-powered scans run only in the Claude Security app on Claude.ai, while the separate Claude Security plugin for Claude Code, now in beta for all Claude Code users, uses whatever models the user's Claude Code account has.

How it fits Anthropic's security push

The update lands amid a broader shift in how AI is used on software security. Vendors have started building on Mythos directly, as with Chainguard's Harness for vulnerability hunting, and large companies are reorganising around the volume of findings; Bank of America brought in 1,000 contractors to patch flaws surfaced with Anthropic's models.

The flood has side effects for maintainers too. The curl project recently said it will take low and medium severity vulnerability reports public to cope with a surge of AI-driven submissions. Fix tracking and threat-model-based prioritisation are Anthropic's answer to the same problem on the defender's side: fewer stale findings and less noise.

Anthropic's announcement includes customer quotes, one of which says the API and MCP additions will be IonQ's product security team's path to further operationalising the tool. The quotes are vendor-selected testimonials.

Caveats and open questions

Anthropic's own product page warns that Claude can make mistakes and that teams should always review proposed patches before applying them, especially on critical systems. The tool is still in beta and limited to Enterprise customers.

Several things are not public. Anthropic has not published false-positive rates, how often continuous scans run in practice, how the risk-based scheduling decides which changes to rescan, or what the credits cover in terms of scan volume. The million-vulnerability figure is not independently verified. Teams connecting repositories through access tokens should also weigh what code and context they are sending to Anthropic's cloud, since the Mythos scans run there rather than locally. AI Lookout has not tested the new version.