PromptBrake
Find AI vulnerabilities before hackers do
4 followers
Find AI vulnerabilities before hackers do
4 followers
Most AI security testing takes weeks and needs experts. We made it stupid simple! Paste your endpoint. We attack it with 60+ real exploits (prompt injection, data leaks, jailbreaks). In a couple of minutes = full security report in plain English. Works for solo devs to enterprise teams. OpenAI, Claude, and Gemini supported. API keys are never stored. Catch vulnerabilities before they catch you.
Hi ProductHunt! 👋 I'm Ammar, creator of PromptBrake.
I built this because I kept watching teams (including mine) ship AI features while secretly hoping nobody would try to break them.
The problem? OWASP docs felt like reading a PhD thesis. Most of us just... shipped and prayed. I literally lost sleep over this.
PromptBrake is what I needed back then: Drop in your AI endpoint (OpenAI, Claude, Gemini, whatever).
We throw 60+ real attacks at it. A couple of minutes later, you get a clear report showing what's vulnerable and how to actually fix it.
No security background required. We catch the stuff that keeps founders up at night - prompt injection, data leaks between users, jailbreaks, and system prompt extraction. The attacks people actually use, not theoretical ones.
Here's my question for you: If you're building with AI right now, what's actually stopping you from security testing? Is it the cost? The complexity? Not knowing what "good enough" even looks like? Genuinely curious. Drop a comment - I'm here all day answering questions.
This is something I've been doing manually for our chatbot SaaS, running jailbreak prompts against the API before each release. Having it automated with CI/CD gates would save a ton of time. Does it support testing custom system prompts, or only the raw API endpoint?
@cuygun - Yes, both are supported. You can scan a raw endpoint, and you can also test with a custom system context via Example Prompt or your exact payload structure via Request Template (as shown). We also support CI/CD gating, so builds can fail automatically when ci_gate.passed is false.
https://promptbrake.com/docs
@promptbrake, Perfect, the request template option is exactly what I need. We have a pretty custom payload structure with context injection. Going to try the CI gate integration this week, appreciate the quick response.
@cuygun - Awesome, that’s exactly the workflow we built for.
For custom payloads, request_template is the right path, and CI gate setup is straightforward once your CI key is in place. Also, there’s a free trial with a couple of scans, so you can validate everything quickly before committing. If you need any help during setup, I’m happy to help.
@cuygun - checking in! How'd the CI integration go? Curious to hear if the request template handled your custom payload structure smoothly.