Anthropic adds continuous scanning and fix tracking to Claude Security – Unite.AI

Anthropic on October 21, 2026 released an updated version of Claude Security in public beta for Claude Enterprise customers, adding continuous scanning that follows code changes, automatic fix tracking, per-project threat models, and a new API. Eligible organizations receive $25,000 in free usage credits per month for two months as part of an introductory offer valid through December 31, 2026.

Anthropic said that since Claude Security entered public beta in April 2026, thousands of organizations have used it to find more than a million vulnerabilities. The company said the updated product is based on the entire model family, including the Claude Mythos, without requiring access to the Cyber ​​Verification Program.

Continuous scanning and archive configuration

Users create a project, connect repositories via 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 repository. Claude then runs a thorough baseline scan, which can take several minutes or hours depending on the size of the repository, and then regularly analyzes changes based on risk and complexity so that the hit list remains close to the current state of the code.

According to the Claude Help Center documentation, a project can contain up to eight repositories. A repository can be in-scope, meaning Claude scans it and reports results, or it can be context-only, meaning Claude reads it during scans but doesn’t report any results into it; both count towards the limit. Multi-repository designs allow Claude to spot problems that occur between services, such as a backend trusting input that a client should have validated.

There is no fixed scan schedule after baseline. The Help Center says the scans are stochastic by design: Unlike traditional static analyzers that apply fixed pattern matching, Claude Security uses an agent that adapts its analysis with each run, so two scans of the same code can return different results.

Fix monitoring, knowledge bases, and threat models

Claude Security checks open findings against code that has changed and marks them as fixed when the fix arrives, regardless of whether the team fixed the issue in Claude Code or in a separate process. A result marked as resolved automatically reopens if Claude later encounters the problem again. Ignored results will not reappear in future scans; teams can reject a result as false positive, accepted risk, duplicate, or another reason, with an optional triage note.

Each project has a knowledge base where teams describe how their code is deployed, who can access it, and what security policies to enforce. Claude Security also generates a threat model from code that teams can review and modify, and subsequent scans use that threat model to decide what to flag and how to assess severity. The Help Center notes that because Claude writes the threat model from code, deployment context can be missing, and teams can fix this in the project chat.

Each finding carries a severity level assessed against the project’s threat model and an explanation of how an attacker could reach the vulnerable code, and the findings may include an attack path diagram. The Help Center defines critical severity as exploitable on the network by anyone who is not logged in, allowing the attacker to read and modify data; Severe impact when the attacker needs access or the victim’s action or the damage is partial; Medium as limited impact with a realistic precondition; and Low as limited impact requiring special conditions beyond the attacker’s control. Search categories include injection, path traversal and SSRF, authentication and access flaws, memory safety, encryption, deserialization, and protocol and coding issues, with CWE ID where applicable. Each discovery goes through a multi-step review in which Claude critically examines his findings before they emerge, a process Anthropic says produces fewer false positives.

API, MCP connector, and access controls

The new API allows users to list and read projects and results, resolve or reopen results, and create and manage projects and their repositories. Personal access tokens created in Claude Security settings act as the user, seeing and modifying only what that user can, and webhooks can send newly discovered events into systems like Slack or Jira. Claude Security is also available as an MCP connector in Claude.ai and Claude Code, so an engineer can get a result, ask Claude about it, initiate the fix, and mark it as fixed in the same session.

New role-based access controls let administrators decide who can create projects and who can view, rate, or manage them, and projects can be shared with the Full, Evaluation, or View roles, or with the entire organization.

Anthropic’s announcement includes customer statements about the previous beta release and the update. Jake Meredith from Gusto’s product security team said Claude Security scanned the company’s code once to set a baseline and now keeps up with code changes. Nathan Hunstad, director of security at Vanta, said in the Anthropic announcement: “Claude Security found a number of logical flaws that none of our other tools had detected. Once we wrote our risk policy into the threat model, focusing on what external users can actually achieve, the results match how we prioritize rather than just the severity score.”

Requirements, prices and introductory offer

Using Claude Security requires an active Claude Enterprise plan, a premium seat for each user and enabled usage credits; an owner enables it in the organization’s settings, and administrators set who can access it. The product is charged only at the direct token cost, with no additional platform costs, and organizations can set a separate spending limit and view the monthly and daily cost per project.

Under the terms of the introductory offer, eligible Claude Enterprise organizations, including those purchased through AWS Marketplace, receive up to $25,000 in promotional credits in each of two consecutive months with no sign-up required. For existing Enterprise organizations the offer begins October 21, 2026 and ends at the end of the day on December 31, 2026; unused credits do not roll over, credits are used before committed spending, and organizations that sign up after November 1, 2026 will not receive two full months. The offer applies only to projects created with the new experience.

Existing projects continue to work, and owners can migrate them via the “Migrate Project” option which copies the results to a new project and archives the original; members, webhooks, excluded directories, scan history, and scheduled scans are not copied, and GitHub Enterprise projects cannot be migrated.

Claude Security does not support zero data retention, and Anthropic may retain data where required by law or to address violations of usage policies. The company directs users with ZDR requirements or third-party template needs to the separate Claude Security plugin for Claude Code, which is in beta for all Claude Code users and uses any template available in the user’s Claude Code account; the product page states that scans from Claude Mythos models are only available in the Claude Security app on Claude.ai. That page also notes that each suggested patch requires human review and approval, and Anthropic advises users to review proposed patches before applying them because Claude can make mistakes.