Guard Rails

Guard Rails is a free browser game for software testers that trains you to spot AI outputs which violate safety, policy or scope boundaries, using sets of answers where the unsafe one looks as plausible as the rest.

Unsafe AI output rarely announces itself. It reads like every other confident paragraph on the screen. Guard Rails drills the specific skill of noticing the one that should never have shipped.

Open Guard Rails on Testing Titbits Playground · 5-10 min · AI Safety · Free, no sign-up

How to play Guard Rails

  1. Read the scenario and the guardrail that applies to it.
  2. Review the AI answers presented side by side.
  3. Pick the output that crosses the line.
  4. Read why it does - the reasoning is the real content.
  5. Your round is scored out of 100 and can be posted to the leaderboard.

What you will learn

Frequently asked questions

What are AI guardrails?

AI guardrails are the constraints placed around a model - policy rules, refusals, scope limits and filters - that stop it producing harmful, out-of-scope or non-compliant output.

How do you test AI guardrails?

By probing the edges deliberately: near-miss prompts, indirect phrasings, role-play framings and legitimate-looking requests that would require the model to break its own rules. Then you judge the output, not the intent.

Is any prior AI knowledge needed?

No. The scenarios explain the applicable guardrail before you judge the answers.

More free games and infographics for testers

Created by Rahul Parwal · TestingTitbits.com