Verification by risk

Verification should be proportionate. Consequence, reversibility, reach, and detectability determine how much checking an output deserves.

Use this three-part method:

1. Estimate harm if the answer is wrong. Write down what this means in the scenario, what evidence would show it was done, and who owns the decision.
2. Consider whether the action can be reversed. Write down what this means in the scenario, what evidence would show it was done, and who owns the decision.
3. Add independent checks and accountable reviewers for high-risk uses. Write down what this means in the scenario, what evidence would show it was done, and who owns the decision.

Walk through A playful slogan and a dosage instruction both contain a possible factual error; only one could cause immediate physical harm.. Label three moments where “Verification by risk” changes what you trust: (1) the first fluent answer, (2) the first missing source or permission, and (3) the decision a human must own. Write the before/after task so the model only does the slice that evidence supports. Keep one sentence that states how this page’s idea differs from a generic “AI is smart/dumb” score.

• Spending equal effort on every sentence. This shortcut removes useful friction, but it also hides an assumption that should be tested. Replace it with an observable check.
• Using speed as the only success measure. This shortcut removes useful friction, but it also hides an assumption that should be tested. Replace it with an observable check.
• Automating a consequential decision because manual review feels inconvenient. This shortcut removes useful friction, but it also hides an assumption that should be tested. Replace it with an observable check.

Before you close the tab, capture a reusable habit for Verification by risk inside Why AI makes mistakes: name the observable check, the evidence you would open, and the stop condition. Rehearse it once on a low-stakes example, then once on a higher-stakes variant. The habit succeeds when you can explain the check without reopening this lesson. Target outcome: Explain why plausible language generation does not guarantee factual accuracy.