Anthropic has expanded its AI-driven vulnerability detection capabilities by launching Claude Security as a native Claude Code plugin, now available in beta for all Claude Code users.
The move brings Anthropic’s reasoning-based security scanning directly into developers’ terminals, allowing teams to scan changes before committing or run full codebase audits using the same Claude inference they already rely on for coding tasks.
Installation follows a straightforward workflow: developers install the plugin from the official Anthropic marketplace using /plugin install claude-security@claude-plugins-official, then activate it in the current session with /reload-plugins to apply changes without restarting.
Claude Security Plugin Uses AI to Find Flaws
Once enabled, running /claude-security triggers the interactive scan menu, which reads the repository first before offering scan-scope options based on file count and relative cost.
Claude Security is engineered to target high-severity vulnerability classes that traditional static analysis tools frequently overlook, including memory corruption, injection flaws, authentication bypasses, and complex logic errors.
Its core differentiator lies in detecting context-dependent vulnerabilities that require reasoning across multiple files rather than isolated pattern signatures.
Under the hood, the plugin maps a repository’s architecture, constructs a threat model, and dispatches specialized researcher agents across the codebase to trace data flows and business logic much like a human security researcher would.
Unlike conventional rule-based scanners that rely on static signature matching an approach prone to high false-positive rates and blind spots for novel logic flaws Claude Security reasons through code semantically.
It reads Git history, traces data flow paths, and interprets business logic to validate whether a suspected flaw is genuinely exploitable before surfacing it.
Confirmed findings then pass through an adversarial review panel that scrutinizes each proposed fix before it reaches the developer, a mechanism designed to suppress noise and improve confidence in generated patches.
The plugin supports both targeted and comprehensive scanning modes. Teams can scope scans to a specific directory or focused area such as an API layer or authentication module rather than analyzing an entire large repository at once, with the scan report explicitly stating what was and wasn’t examined.
Scans can also target a pull request diff, a single commit, or the uncommitted changes on a branch, giving developers a pre-commit safety check.
Scheduled scans provide ongoing coverage without requiring manual initiation, an addition Anthropic introduced to support continuous security posture monitoring.
Claude Security is designed to plug into existing security workflows rather than replace them. Findings can be routed to Slack, Jira, or any ticketing system via webhooks, or exported as CSV or Markdown for audit trails and compliance tracking.
Notably, documented dismissal decisions persist and carry forward across sessions, ensuring triage history remains visible to the entire team wherever they work.
The plugin builds on Anthropic’s earlier Claude Code Security research preview, launched in February 2026, which brought similar codebase-scanning capabilities to Claude.ai’s web interface for Enterprise customers before this broader terminal-based rollout.
Cut SOC investigation blind spots and contain threats earlier to reduce response costs and business disruption with ANY.RUN.
