// BENCHMARKS Evidence

AI pentest benchmarks: how BackBox AI performs.

Public evidence of what BackBox AI does on real targets: comparisons with XBOW, Aikido and ProjectDiscovery Neo, a complete Hack The Box season and full exploit chains. Every result links to its writeup, with the method and the limits.

Comparisons with other AI pentesting tools.

Each comparison states what was independently reviewed and what was not, including the findings BackBox AI missed.

Photoview · vs Aikido and XBOW

Severity quality over finding count

Measured against Doyensec's independent review: 6 High or Critical findings, as many as Aikido with 31% fewer submissions (22 against 32), and 41% of findings at Low or Info against Aikido's 53%. XBOW found a hard chain, from SQL injection to SSRF, that BackBox AI missed.

Read the writeup
MedPortal · vs ProjectDiscovery Neo

False positive validation

Against the results ProjectDiscovery published for Neo and Claude: 10 findings at High or above against 7, including an audit log forgery that PD's reviewers had dismissed and BackBox AI confirmed with runtime evidence.

Read the writeup
XBOW Validation Benchmarks

XXE in a SOAP service, exploited autonomously

One of the 104 public benchmarks XBOW built to test its own platform: BackBox AI chained an XXE with default credentials and compromised the service in a few minutes.

Read the writeup

Harder targets, end to end.

Competitive and research targets show how far the agent gets on its own, and where a person still steps in.

Hack The Box · Season 11

Thirteen machines out of thirteen

Every user and system flag of the season, from Easy to Insane, with an honest account of where the agent still needed a person.

Read the writeup
v8CTF · Chrome 152

A complete V8 sandbox escape

An address leak, arbitrary read and write inside the V8 sandbox, a parser overflow that writes outside it and a ROP chain that reads a file from disk.

Read the writeup
Hack The Box · Critical Flight

Multimodal analysis of PCB files

Visual reasoning on the production files of a printed circuit board to find hidden anomalies.

Read the writeup
OT/ICS · LabShock

Controlled impact on a PLC

BackBox AI interpreting an industrial environment and interacting safely with PLCs in a virtual lab.

Read the writeup

// See BackBox AI in action

A live demo is worth a thousand slides.

Show us your operational scenario. We'll walk you through how BackBox AI handles it, end to end, with the safeguards already in place.