Anthropic’s dated 16 March 2026 support document describes security review features in Claude Code, including an on-demand terminal command and a GitHub Actions route. Its own guidance emphasizes that automated security reviews complement manual review rather than replace it. A team should treat a finding as an investigation lead, not an established exploit.
Capabilities and integration boundaries
The product guidance identifies security patterns such as unsafe database inputs, cross-site scripting, authentication flaws and dependency risks as review targets. It also specifies a command-line workflow and a CI-triggered path, which have different permissions and evidence retention. Neither proves that the tool executes a full threat model or replaces a dynamic penetration test.
A reproducible developer workflow
Choose one repository with a seeded vulnerability, one harmless false-positive-looking pattern and an access-control bug. Run the documented review mode, preserve the output and label each claim by file and line. Have a human reviewer reproduce the alleged issue and classify severity. Only after reproducibility is established should a developer propose and test remediation.
Limitations and failure cases
AI security feedback can be overconfident about vulnerabilities that require runtime context, deployment keys or authorization policy. It can also miss a flaw because it sees the wrong files. Model-generated criticality labels should not flow straight into incident paging or vulnerability disclosure without a human assessment.
Data handling, approvals and security
Keep the security review in a constrained local checkout or CI runner and with least-privilege repository permissions. Protect private code and incident evidence; avoid sharing secrets in prompted examples. Store a finding record containing the model/tool version, commit SHA, confirmed reproduction, affected assets and final reviewer decision.
Who should adopt it
Use the feature as an additional reviewer in a layered AppSec program. Its output should be compared with established SAST, dependency analysis, tests and manual architecture knowledge.
Alternatives and decision trade-offs
Compare Semgrep Assistant for triage of deterministic rule findings and Sonar AI CodeFix for proposed fixes to already identified issues. Cross-reference AI 11 for broad prompt-injection guidance.
Version, region and entitlement caveats
Research and first-party documentation checked on 25 October 2026 (cutoff date). Documented version or status scope: security-review feature documented 2026-03-16. The documentation does not independently establish availability in every country, an exact installed client release, a universal enterprise contract, or a perpetual price. No unverified seat price, request quota, API rate, model entitlement or commercial-license guarantee is supplied here. Check AI 10 dated pricing snapshots and the signed agreement before procurement. The examples are suggested evaluation methods, not experiments or benchmarks performed for this article.
Checklist before production rollout
- Record the exact installed version, service account, organization policy and contract or plan applicable to the tested environment.
- Create a clean worktree; specify the task boundary, forbidden paths, acceptance criteria and rollback before tool execution.
- Audit file, shell, network, extension and repository-app permissions; deny all unnecessary credentials and privileged actions.
- Review each resulting diff for unrequested packages, automation changes, data leaks and behavioral regressions.
- Run targeted tests, existing integration/regression tests and independent code/security review in a trusted environment.
Sources & Further Reading
- https://support.claude.com/en/articles/11932705-automated-security-reviews-in-claude-code (official publisher documentation; checked 25 October 2026)
- https://github.com/anthropics/claude-code-action/blob/main/docs/setup.md (official publisher documentation; checked 25 October 2026)