Claude Security Plugin: AppSec Moves Into the Claude Code Loop
Anthropic put the Claude Security plugin for Claude Code in beta—diff or full-codebase vulnerability scans from the terminal on existing Claude inference. First-party plugin on the developer’s critical path; treat ‘finds what SAST misses’ as aspirational until independent studies catch up.
Coding agents are writing more production code; security review has to move into the agent loop. A first-party Claude Code Security plugin puts Anthropic on the developer’s critical path—the same week Google’s limited Flash Cyber and OpenAI’s cyber-eval breach news keep dual-use discovery in headlines.
Anthropic announced that the Claude Security plugin for Claude Code is available in beta. Developers can scan changes for vulnerabilities before commit or run a full codebase scan from the terminal on the Claude inference they already run. Official install path: /plugin install claude-security@claude-plugins-official. Product surface: claude.com/product/claude-security.
What shipped
Per Anthropic’s product account (@claudeai) and same-day technical coverage:
| Capability | Detail |
|---|---|
| Scope | Diff / pre-commit scan or full codebase scan |
| Runtime | Inside Claude Code terminal sessions |
| Agents | Multi-agent vulnerability analysis (reported six-phase scan flow in secondary technical write-ups) |
| Install | Official plugin marketplace: claude-security@claude-plugins-official |
| Inference | Uses the Claude session the developer already has—no separate scanner SaaS handoff required for the basic path |
This is a Claude Code plugin expansion of Anthropic’s broader Claude Security product line. Earlier in 2026 Anthropic had put Claude Security in enterprise-facing public beta (codebase scan + proposed patches); the July 22 move is the terminal plugin packaging for day-to-day Claude Code workflows.
Context in the security stack
| Surface | Role (2026 landscape) |
|---|---|
| Claude Security (enterprise beta) | Org-scale codebase scan + fix suggestions |
| Claude Security plugin (Jul 22) | In-terminal Claude Code workflow |
Automated /security-review / GitHub Action |
Earlier PR/ad-hoc review path in Claude Code |
Secondary reports note multi-agent validation of findings before suggesting patches—important because raw LLM “security review” is noisy. Treat marketing “finds what SAST misses” claims as aspirational until independent red-team studies catch up; still, shipping a named official plugin is a concrete product line, not a blog tip.
Limits
- Beta; false-positive rates and fix quality are not independently scored here.
- Six-phase agent flow details are secondary technical write-ups.
- Enterprise Claude Security portal vs plugin packaging overlap is still settling.