Module 4 · 2:25

When the Guardrail Catches You

Redacted, blocked, or warned: what each means and why nobody is reading your chats.

What to remember

The safety net has three moves: redact masks and sends, warn hands you the decision, block stops the never-paste classes entirely.
Tags like [NAME] and [CREDENTIAL] are not error codes; they tell you exactly what was caught and why.
The AI's answer works with placeholders; the last step is swapping the real details back in on your side, after it comes back.
If the net catches the wrong thing, one click flags it, and that helps everyone.
No human is reading your chats; what is kept is a compliance record, and getting caught is the system working, not you failing.

Quick check

Three questions, instant feedback. Two right completes the module.

1.The safety net raises a warning, not a block, on your paste. What does that mean?

2.The AI answered your redacted question using placeholders like [NAME]. What is the last step nobody teaches?

3.With a safety net running, what does your company actually keep from your AI chats?

Transcript

0:01: The safety net caught a paste. Today, the full tour.
0:17: Three moves: redact, warn, block.
0:25: Sensitive spans become labeled chips. The question goes through anyway.
0:42: Tags are not error codes. They say what was caught and why.
0:54: The answer reuses placeholders. Swap the real details back in after.
1:10: A warning means you decide. A block means never, at any company.
1:29: Caught the wrong thing? Flag it, that helps everyone.
1:43: No human reads your chats. What is kept is a compliance record.
2:02: Redacted: safe. Warned: you choose. Blocked: never. Keep using AI.

Keep going

Next up: Trust, Verify, and AI-Made Deception

The whole course is free and login-free. IT and security teams get the rollout kit with a quiz pack and an EU AI Act evidence checklist.

Start module 5 All modules