Berean Labs – Free AI-powered penetration testing for web apps

Hey HN,

I'm sharing Berean Labs (https://bereanlabs.com), a free, autonomous AI penetration testing tool designed to catch client-side vulnerabilities, exposed secrets, and misconfigurations in your web apps before attackers do.

The Problem Traditional DAST/SAST tools are often expensive, hard to configure, or generate massive amounts of false positives. They also sometimes lack the semantic understanding to realize that an exposed <!-- AWS_KEY=... --> comment or a specific combination of DOM sinks is actually a critical vulnerability. I wanted to build a tool that acts more like a junior red-teamer looking at your frontend code.

How it works under the hood 1. Domain Verification: To prevent abuse and random scanning, you first verify domain ownership via a DNS TXT record. 2. Safe Fetching: The Node.js backend fetches your target's HTML. I implemented strict SSRF protections here—it resolves DNS and explicitly blocks private/local IPs and localhost routing before fetching. 3. Attack Surface Extraction: Using Cheerio, the backend parses the DOM to extract a highly condensed "attack surface summary" to fit into the LLM context window. This includes forms, input fields, external script sources, suspicious inline scripts (e.g., eval, innerHTML), inline event handlers, and HTML comments. 4. AI Analysis: This sanitized context is fed into a specialized model (powered by Abliteration.ai) via a strict red-team system prompt. 5. Structured Reporting: The model enforces a JSON schema to return vulnerabilities ranked by severity, complete with CVSS scores, affected code snippets, and remediation steps.

It's completely free to use. I built this primarily to see how well current LLMs can perform context-aware security auditing on raw client-side output when given the right constraints.

I'd love for you to try it on your own domains and let me know what you think. Does it catch things your standard linters/scanners miss? Are there false positives that annoy you?

Check it out at: https://bereanlabs.com

Feedback, questions, and roasts of the architecture are highly welcome!