Anthropic's Claude Security Plugin Is Now in Public Beta for All Claude Code Users

Anthropic has opened Claude Security to all Claude Code users in public beta. The tool, which launched in early 2026 as a limited research preview for Enterprise and Team accounts, is now accessible to anyone using Claude Code.

What It Does

Claude Security integrates vulnerability scanning directly into the Claude Code terminal workflow. The entry point is a /claude-security command that opens three options: scan recent uncommitted changes, run a full codebase scan, or run continuous monitoring on a directory.

The scanner focuses on vulnerabilities that pattern-matching tools typically miss: injection flaws, authentication bypasses, memory corruption, and complex logic errors that span multiple files. Rather than checking code against a signature database, it analyzes how data flows through an application and how components interact.

The system runs as a pipeline of agents. One maps the codebase architecture, another identifies threats, a third independently verifies each finding to filter out false positives, and a final agent generates fix suggestions matched to the surrounding code style. Nothing gets auto-committed. Developers review and approve every suggested patch.

Three Layers of Review

According to Anthropic’s description, the plugin operates at three different points in a developer’s workflow:

A lightweight pattern-matching pass while you edit
A deeper check after each Claude Code interaction
A broader agent-led review at commit time

The commit-time scan is the most thorough, re-examining recent changes with the full multi-agent pipeline before any code leaves your machine.

Integration

Findings can be exported to Slack, Jira, CSV, or Markdown through webhook integrations. The plugin also supports scheduling periodic scans across a directory.

Organization admins enable it through the admin console at claude.ai/admin-settings/claude-code. There’s no additional pricing listed — it’s available to all Claude Code users in beta.

Background

Anthropic first announced Claude Code Security in February 2026, when a research version of the tool surfaced more than 500 previously unknown vulnerabilities in open-source codebases. That version was available only to Enterprise and Team accounts under a limited preview. This public beta is the broader rollout.

The main competitive context: GitHub Copilot has its own vulnerability detection (Autofix), and the security scan feature set has become a standard part of AI coding tool positioning. What Anthropic is emphasizing here is the multi-agent verification step — the claim that the system argues against its own findings to reduce noise before surfacing results to developers.

Full details at claude.ai/code.