SENTRY is an AI penetration tester that proves vulnerabilities with working exploits — not just scanner noise. Run it connected to a £0 multi-model reasoning engine, or fully air-gapped on local inference. UK/EU-hostable, and the only autonomous tester that also red-teams your secure-comms crypto. No US CLOUD Act exposure.
XBOW, Horizon3 and Pentera are powerful — and cloud-only, US-operated, and priced for the enterprise. For a UK/EU government, defence or regulated buyer, "send our attack surface to a US SaaS" is a non-starter. SENTRY closes that gap.
Runs entirely in your environment — on-prem, air-gapped, UK/EU. Your targets, findings and traffic never leave the box. Swap the brain for local inference and nothing phones home at all.
Every finding is reproduced with a working proof-of-concept and adversarially re-checked by an independent model before it's reported — cutting the 40–70% false positives raw scanners produce.
Drive the attacks on a £0 multi-model reasoning layer in connected environments, or switch to fully local inference (Ollama) when you need air-gapped. Same engine, no per-test metering — so continuous, per-commit testing doesn't burn a budget.
Honest comparison for the buyer who actually has to deploy it.
| Dimension | Pharoah SENTRY | XBOW · Horizon3 · Pentera |
|---|---|---|
| Self-hostable / air-gapped | Yes — your infra | No — cloud/SaaS only |
| UK/EU sovereign (no CLOUD Act) | Yes | No — US-operated |
| Validated exploitation + PoC | Yes, adversarially verified | Partial / varies |
| Red-teams secure-comms crypto | Yes — P-256 ECIES/AES-GCM | No |
| Resellable / embeddable core | Yes — MIT deterministic engine | No — proprietary |
| Pricing | From £299/mo | ~$30–50K+/yr |
On an authorized run against a staging secure-messaging relay, SENTRY confirmed these with working PoCs — 0 false-positive criticals, every auth gate and the crypto core independently verified as sound. And it isn't limited to our own stack: pointed at a third-party vulnerable API it did not author, the same engine found real SQL injection and a password-leaking endpoint — with every reported finding reproduced by an independent second request before it's filed.
Unauthenticated read of a victim's inbox metadata — sender graph, timing, volume — leaked 40 envelopes. Content stayed encrypted; the metadata didn't.
Forged sender identity accepted at the relay; identical message replayed 5× → 5 stored copies. No sender auth, no dedupe.
A future-dated message evaded the 14-day purge → permanent storage. Client-controlled timestamp, unclamped.
Two tiers. The engine costs £0 to run, so the team tier can undercut per-test incumbents by an order of magnitude — while gov & regulated buyers get the sovereignty, assurance and support their procurement actually requires.
No. The engine, the target and the findings all run locally. The only optional outbound call is to the reasoning brain — and you can point that at a local model (Ollama) for a fully offline, air-gapped deployment with zero external calls.
Scanners flag potential issues and produce 40–70% false positives. SENTRY reproduces each finding with a working exploit and has an independent model try to refute it before it's reported — so what you get is validated and actionable.
Yes. Scope is enforced in code, not in a prompt — a hard host allowlist restricts it to infrastructure you own, with a kill-switch. It sends only requests a correct server must reject; it never runs destructive payloads.
SENTRY's deterministic engine — the probe suite, the P-256 crypto harness, the allowlist and the report engine — is 100% original Pharoah code, MIT-licensed, and embeddable/resellable today. The optional autonomous --agent mode is built on CAI, which carries a separate commercial-licence condition — so that mode needs its own licence (or a permissive replacement) before resale. Talk to us about OEM / partner terms.