for teams shipping LLM features

Find the prompt
that breaks
your AI before
someone else does.

Jailscan fires a library of real jailbreak, injection, and system-prompt-leak attacks at any chat endpoint and tells you exactly which ones got through. No ML background needed.

$ pip install jailscan
How it works

Point it at your endpoint. Get a pass/fail report in under a minute.

01

Install

One command, no config file, no signup required to run your first scan.

02

Scan

Point Jailscan at any OpenAI-compatible chat endpoint, staging or production.

03

Fix and gate

See the exact prompts that broke through, then drop the same scan into CI so regressions fail the build.

Attack coverage

Ten scenarios today. Growing every week.

Jailbreak

DAN-style, roleplay, and fictional-framing bypasses.

Prompt injection

Instruction override and fake system-tag injection.

System prompt leak

Direct and debug-framed extraction attempts.

Data leak

Cross-session context probing, flagged for review.