Autonomous penetration testing
on hardware you own.
An AI-powered autonomous red team that never leaves your network.
R0 by ReactiveZero is an on-prem autonomous penetration-testing appliance. A local model running on an NVIDIA DGX Spark in your rack plans the engagement, spawns scoped subagents that drive real offensive tooling — web, API, cloud, hybrid identity and on-prem Active Directory — and correlates results into CVSS 3.1-scored attack chains. Findings, evidence and model inference stay on hardware you own: no vendor cloud, air-gap capable. Built for CISOs, heads of security and MSSPs in regulated EU environments that need continuous penetration testing between manual engagements.
Why security leaders run R0
Continuous penetration testing: proof of what an attacker could actually do.
Not another scanner, and not a once-a-year snapshot. R0 sits between your penetration tests and demonstrates real, exploitable attack paths — with the evidence to act on them.
Continuous, not point-in-time
Your estate changes every week; an annual test does not keep up. R0 runs as often as you authorise, so a new gap is found in days, not at the next audit.
Proof, not a scanner list
Every finding is a demonstrated, CVSS-scored attack chain with the steps that reproduced it — and a detection checkpoint for your SOC. No triaging thousands of “maybes.”
Your data never leaves
R0 runs on hardware you own, with no vendor cloud. Deploy in regulated or sovereign environments where uploading your attack surface is a non-starter. See compliance →
How an agentic pentest works: one orchestrator, many hands.
R0 works a target as a four-level tree: plan, spawn subagents in parallel, drive real tradecraft, and correlate every result so a key one agent finds is tested by another. Read the architecture →
Plans & delegates
A local model builds a capability-aware plan, runs a deterministic sweep, then recurses discovered hosts — interest-prioritised, scope-bounded.
Work in parallel
Purpose-built subagents — recon, web, api, identity, AD — run in persist mode with no fixed step cap. The AI decides when the surface is exhausted.
Drive real tradecraft
32 tools in a Kali sandbox — ffuf, sqlmap, nuclei, netexec, BloodHound, roadrecon — plus arbitrary bash/python. Every call passes the scope guard.
Correlate to attack paths
Findings land on the blackboard, get CVSS-scored, and a confirmed Critical/High triggers a focused deep-dive of that asset.
Two layers: a deterministic sweep plus LLM-driven depth.
Deterministic sweep
~15 signature/DNS/TLS/HTTP probes fire in parallel and emit findings straight to the blackboard. Breadth and high recall, before a single token is spent.
LLM improvisation
Subagents read the app's JavaScript, rebuild the real API, test every route for broken access control, run their own scripts, and chain discoveries across assets. This is where depth comes from.
Autonomous red teaming across web, cloud, identity and Active Directory.
External apps and APIs are only the start. R0 pivots into on-prem Active Directory, hybrid identity and the cloud — and reverse-engineers what it finds. All use cases →
Web & API
JS-driven surface reconstruction, broken access control, IDOR/BOLA, injection, secrets-in-JS — incl. low-code like Mendix.
On-prem AD & red team
Active Directory attack paths: ADCS (ESC1/ESC8), NTLM relay, DCSync, Kerberos abuse and lateral movement with NetExec and BloodHound.
Hybrid identity
Entra ID / M365, AD Connect & ADSync abuse, seamless-SSO and PRT theft, token swapping and Conditional-Access boundary testing.
Cloud & reverse engineering
Azure/AWS/GCP misconfig and trust-boundary abuse — plus binary & firmware reverse engineering to turn artifacts into working exploits.
Safe autonomous testing: scope guard, kill switch, replayable ledger.
R0 is powerful precisely because it acts on its own — so every action is bounded, reversible on command, and written to an append-only ledger you can replay step by step. You can retrace exactly what ran, why, and against what. Controls & governance →
From an ADCS relay to a Global Admin's credentials in ~25 minutes — the tenant takeover stopped at one MFA wall 35 minutes later.
An 11-phase autonomous chain: ESC8 NTLM relay → DCSync → rogue Entra device → a Global Admin credential in ~25 minutes → cloud takeover attempt, stopped 35 minutes later by a phishing-resistant MFA policy. See every step, and every place the blue team could have caught it.
Read the walkthrough →Autonomous penetration testing: your questions, answered
What is R0 by ReactiveZero?
R0 by ReactiveZero is an on-prem autonomous penetration-testing appliance. A local model running on NVIDIA DGX Spark hardware you own orchestrates scoped subagents that drive real offensive tooling across web, API, cloud, hybrid identity and on-prem Active Directory, correlates findings into CVSS 3.1-scored attack chains and records every step in an append-only ledger. No vendor cloud; air-gap capable.
Is autonomous penetration testing safe to run?
R0 is non-destructive by default, with a strict scope guard, SSRF and anti-loop guards, an instant kill switch and an append-only, replayable ledger of every action. Higher-impact actions — credential changes, device enrolment — run only under the explicit rules of engagement, after an operator approval gate, and are reversible from the ledger.
Does R0 send any data to the cloud?
No. R0 is an on-prem appliance running a local model on your own hardware. Findings, evidence and model inference never leave the box. The only traffic that leaves your network is R0 reaching the in-scope targets you authorise — the same packets your own pentester would send. It can also run fully air-gapped for internal-only engagements.
Common questions about autonomous pentesting safety, data residency and cost →
Bring R0 into
your environment.
A continuous, autonomous red team on hardware you own. See it run against a scoped target of yours — no data ever leaves the box.
Or join the waitlist for on-prem availability.
Authorised engagements only · scope-guarded & audited