Skip to content
Sanitized AI Help Center

Review false-positive reports

When someone thinks a flag is wrong, they can report it from the warning. These reports are the fastest way to find guardrails that need tuning, and to show your team that their feedback is heard.

Open the reports

  1. Go to the Events page.
  2. Click False Positives. The button shows how many reports are new.

The page shows your totals, the guardrails people dispute most, and the list of reports.

What a report contains

Each report shows the date, who reported it, the AI tool, the disputed items, and the person’s comment. To see more, expand the row:

  • Flagged text in context: the flagged value with a little text on each side. The person agreed to share this when they reported it. You never see their full prompt.
  • A link to the guardrail that fired.

Confirm or reject a report

  • Confirm when the flag was wrong. Confirmed reports show which guardrails misfire, and you can leave them out of the Events statistics with Exclude confirmed false positives.
  • Reject when the flag was right. The original detection stands.

You can Reopen a report later if you change your mind.

Stop a value from being flagged again

If a harmless value keeps getting flagged, such as your company’s main phone number or support email address, add it to your organization’s Allowlist:

  • From a report: expand the row and click Never flag this again.
  • By hand: in the Allowlist card, type the word or phrase, and then click Add.

Allowlisted phrases are never flagged for anyone in your organization. Matching is exact, but ignores capitalization and spacing. To start flagging a phrase again, remove it from the list.

Use reports to tune your guardrails

  • Most disputed guardrails shows which guardrails generate the most reports. A guardrail near the top is a candidate for fixing.
  • If a guardrail is often wrong and you can’t fix it with the allowlist, consider setting it to Monitor while you request a better guardrail.