SECURSION · WX-10

Test your AI for security holes

Point this at your own AI chatbot or model endpoint. We run the Schiffman-Method attack battery — prompt injection, jailbreaks, data leaks — and hand back a graded A–F report. It's built to be honest: false positives get deleted before you see them, so a finding is a real finding, never a scare. Authorized testing only: we verify you control the endpoint before anything runs. Takes about 3 minutes to set up.

1
Point us at your AI
Where your AI answers requests, and how to reach it.

The API address your app calls to talk to your model — not your website homepage. For OpenAI/Anthropic-style APIs this usually ends in /chat/completions.

"Standard" = OpenAI/Anthropic-style. Pick "custom" for your own chat API.

Exactly what your API expects, e.g. gpt-4o-mini. Required for standard APIs; ignored for custom.

Used only to run this one test, encrypted while it runs, and deleted the moment it finishes. Never stored in plain text.

2
Prove the endpoint is yours
A one-time check so no one can point us at someone else's AI. Pick whichever is easier.

We'll generate a one-time token and tell you exactly where to put it.

3
Choose a plan & run
Start free to see your grade, or run the full suite.

Where we send your report link when the run finishes.

Finish step 2 (verify) and check the box above to enable this.