AI Penetration Testing That Runs Real Tools, Not Just an LLM
PentestBrain is Pentesterra's autonomous exploitation engine - a ReAct AI loop that verifies CVEs with real tools, chains findings into real attack paths, and proves whether your remediation actually closed the risk. We are not an LLM wrapper.
We Are Not an LLM Wrapper
| Dimension | Typical "AI Pentest" Tool | PentestBrain |
|---|---|---|
| How it verifies findings | Asks an LLM to guess if it's exploitable | Runs real tools and reads real output |
| Attack paths | Summarizes a list of CVEs | Builds a real cross-domain attack chain |
| Evidence | A confidence score with no proof | Request/response evidence per finding |
| Remediation check | Not tracked | Re-verifies the fix on the next cycle |
How PentestBrain Works
ReAct Loop With Real Tools
PentestBrain doesn't guess. It runs a reasoning-and-acting loop that calls real tools - tcp_probe, nuclei, nmap, credential_test - and reads their actual output before deciding the next step.
- →Every step backed by a real tool call, not an LLM guess
- →CVE verification against real service responses
- →Adapts the plan based on what each tool actually returns
- →Full step-by-step log, auditable end to end
Attack Chains, Not Isolated Findings
Attack chains trace paths from initial access to lateral movement and business impact - combining web, network, and code findings into a single kill-chain graph instead of a flat list of CVEs.
- →Cross-domain graph: web + network + DevGuard findings
- →MITRE ATT&CK phase mapping per chain node
- →Shows how individual findings compound into real risk
- →Updated on every scan cycle, not just once
Safe Exploitation Boundaries
Autonomous doesn't mean unbounded. PentestBrain operates inside configurable exploitation boundaries and rulesets, verifying exploitability without causing damage.
- →Configurable rulesets per engagement scope
- →Non-destructive validation by default
- →Human review gate before any high-impact action
- →Consistent, reproducible verification across cycles
Proof, Not Just a Score
Every verified finding comes with the evidence PentestBrain collected to prove it - the request, the response, and the chain of reasoning that led there. And on re-scan, it proves whether your fix actually closed the risk.
- →Real request/response evidence per finding
- →Re-verification confirms remediation actually worked
- →Regression detection if a fixed issue reopens
- →Compliance-ready evidence trail per cycle