A guardrails library is a defense. A scan is proof it works.
Open-source guardrails libraries add input/output filtering, PII redaction and jailbreak-pattern matching to your chatbot. They're a real, useful layer. But installing one is not the same claim as knowing it actually holds against your specific bot, your specific prompts and your specific data. Here's the honest comparison.
Check your inbox!
Your Starter Map is on its way. If it doesn't arrive in a minute, check spam.
What each one actually does.
These aren't competitors — most hardened chatbots eventually use both. The question is which one tells you whether the other is working.
| Guardrails library | AI Sec Tester scan | |
|---|---|---|
| What it is | Code you install and configure — a defensive layer that filters inputs/outputs at runtime | A one-time scoped check that probes your live chatbot and reports what it finds |
| What it tells you | Nothing by itself — a library installed is not a library configured correctly, or tuned to your bot's actual attack surface | Whether specific attacks succeed or fail against your bot, right now, with evidence |
| Ongoing effort | You maintain rules, update patterns, tune false-positive rates yourself | None — it's a point-in-time check, not a running system |
| Covers architecture-level risk | No — it operates at the request layer, same boundary as a scan | Advisory review only, on the Advanced tier — still not a substitute for an internal review |
| Price | Free/open-source, but your time to install, configure and maintain isn't free | $47 or $197, one-time |
| Best used | As the standing defense, installed and tuned over time | To verify the defense actually catches what it's supposed to — before and after you change anything |
The honest answer: a library without verification is a claim, not a control.
Guardrails libraries are genuinely useful — pattern-matching known jailbreak phrasings, redacting PII before it leaves your system, filtering obviously unsafe outputs. But "we installed a guardrails library" is a statement about what you did, not evidence of what actually happens when someone tries the debug-log reframe or an indirect injection buried in a document your bot summarizes. Default configs miss real attacks constantly, and a library tuned for one use case can be silently ineffective in another.
A scan doesn't replace the library — it tells you whether the library (or your custom prompt hardening, or both) is holding against the specific categories in the OWASP LLM Top 10. Run the free DIY version first if you haven't: 7 test prompts, 15 minutes, no purchase required.
If you haven't installed anything yet
Start with the step-by-step hardening order — system-prompt constraints and output validation catch a surprising amount before you need any third-party library at all.
Should I use a guardrails library instead of getting a scan?
They answer different questions. The library is the defense; the scan tells you if the defense is actually working against real attack patterns. Most hardened setups use both.
If I already installed a guardrails library, am I covered?
Not automatically — default configs and untuned rule sets miss real attacks. Run the free DIY test against your bot with the library active to see whether it's actually catching anything.
Does a scan replace the need for a guardrails library?
No — a scan is a point-in-time check, not a running defense. If you want ongoing filtering, you still need something like a guardrails library or your own prompt/output hardening in place.
Find out if your defenses actually hold.
Normal $47 or Advanced $197, one-time. No charge until your request is reviewed and approved.