Sentinel — Autonomous AI penetration testing with real proof

Autonomous penetration testing with live exploit proof and reports auditors accept.

Real findings, with proof, that auditors accept.

The full OWASP surface: Access control, injection, auth, SSRF, logic flaws and more — each with a specialist agent.
Multi-persona authorization: Every endpoint called as every identity. A finding is a disagreement between identities, never one response.
Proof, not guesses: Code root cause plus live proof, three-vote verification and AutoTriage keep the list high-signal.
Rules of engagement: A reasoning browser agent enters destructive flows to prove reachability — and stops before the final commit.
Closed AutoFix loop: Finding to patch to PR to re-test: verified as solved, or reopened.
Compliance-ready reports: CVSS 3.1, OWASP and CWE, mapped to SOC 2, ISO 27001, PCI DSS v4.0, HIPAA and NIST.

Scope and log in: Add your staging host and test personas — TOTP and email codes handled.
Recon and static: Crawl, port scan, SAST and SCA across code and dependencies.
Attack and validate: Specialist agents probe, prove and vote on every candidate.
Report, fix, re-test: Hand over the report, merge the fix, confirm it’s solved.

Scanners: nmap, Semgrep, Trivy, Nuclei, OWASP ZAP
Models: Claude Code CLI, OpenAI, Anthropic, Vertex / Gemini, Bedrock, LiteLLM
CI: GitHub Action security gate and inbound PR reviews
Safety: Egress allowlist, sandboxed execution, non-destructive by design
Status: Closed beta