# What are false positives and false negatives in AI? > A false positive is a flag on something harmless. A false negative is a miss on something real. Every AI sorter, scam check or filter makes both kinds. Source: https://learn.ynteractive.com/content/glossary/false-positive · Updated 2026-10-21 · Free from Stepthrough (https://learn.ynteractive.com) [AI basics](https://learn.ynteractive.com/content/glossary#topic-ai-basics) · [AI glossary](https://learn.ynteractive.com/content/glossary) Quick answer A false positive is when a check flags something that is actually fine, like labeling a newsletter as a sales lead. A false negative is when it misses something real, like letting a fake bank-change email through. Every AI filter, sorter or scam check makes both kinds of mistake, so you decide which one you can least afford. Last updated October 21, 2026 [Take the free module →](https://learn.ynteractive.com/courses/ai-security/4)Module 4 of the free [AI Security for Small Business](https://learn.ynteractive.com/courses/ai-security) course, about 7 minutes. No account needed. Updated October 21, 2026 · By [Robert Breen](https://learn.ynteractive.com/content/about) ## Why it matters for a small business False positives waste time and wear down trust: if the scam alert fires on every vendor email, staff start ignoring it. False negatives are quieter and often more expensive: one missed fraudulent payment can cost more than a year of false alarms. Tuning a tool to catch more usually raises false alarms, and tuning it to be quieter usually raises misses. That is why a sensible process never relies on the AI check alone for high-stakes calls. A scam or injection check is a helpful extra pair of eyes, but the rule that actually protects you, such as calling the vendor back on a number you already have, works whether the AI caught the problem or not. ## In a real lesson: Stop a Fake Vendor Bank-Change Request Before You Pay using Claude In [Stop a Fake Vendor Bank-Change Request Before You Pay](https://learn.ynteractive.com/content/stop-a-fake-vendor-bank-change-request-before-you-pay), Priya pastes a vendor's "we changed banks, pay today" email and a voicemail transcript into a new Claude chat. Claude calls it a textbook vendor-impersonation scam and lists red flags. That is a true positive: a real scam, correctly flagged. The lesson doesn't stop there, because a false negative is always possible. Its three rules, call back on a number you already have, a second person approves any bank change, and slow down when it is urgent, still work on the day the AI misses one. The [prompt injection lesson](https://learn.ynteractive.com/content/catch-a-poisoned-email-before-your-ai-assistant-obeys-it) says the same: Claude caught the hidden line in the test, but a different email or assistant may not. [Try this lesson free](https://learn.ynteractive.com/modules/ai-payment-scams) or [read the step-by-step guide](https://learn.ynteractive.com/content/stop-a-fake-vendor-bank-change-request-before-you-pay). ## Common confusions ### False positive vs false negative Positive means the check said "yes, flag it". A false positive flagged something harmless. A false negative said "fine" about something that wasn't. ### False positive vs hallucination A hallucination is invented content. A false positive is a wrong yes from a check or classifier, even if everything it wrote is real. ## Tips - Decide which mistake costs more for each check before you tune it. - Keep a human step, like a callback, that works even when the AI misses something. - Log false alarms too; if there are many, people will stop reading the alerts. ## Related terms [Precision and recall](https://learn.ynteractive.com/content/glossary/precision-and-recall) · [Phishing](https://learn.ynteractive.com/content/glossary/phishing) · [Prompt injection](https://learn.ynteractive.com/content/glossary/prompt-injection) · [Deepfake](https://learn.ynteractive.com/content/glossary/deepfake) · [Text classification](https://learn.ynteractive.com/content/glossary/text-classification) · [Human in the loop](https://learn.ynteractive.com/content/glossary/human-in-the-loop) ## More AI basics terms [Fine-tuning](https://learn.ynteractive.com/content/glossary/fine-tuning) · [Foundation model](https://learn.ynteractive.com/content/glossary/foundation-model) · [Frontier model](https://learn.ynteractive.com/content/glossary/frontier-model) · [Generative AI](https://learn.ynteractive.com/content/glossary/generative-ai) ## Where you use it: free lessons - [Stop a Fake Vendor Bank-Change Request Before You Pay using Claude](https://learn.ynteractive.com/content/stop-a-fake-vendor-bank-change-request-before-you-pay) (Claude, 7 min) - [Gmail AI Triage: Auto-Label Every Email and Draft Replies](https://learn.ynteractive.com/content/gmail-ai-triage-labels-drafts-n8n) (n8n, 18 min) - [Catch a Poisoned Email Before Your AI Assistant Obeys It using Claude](https://learn.ynteractive.com/content/catch-a-poisoned-email-before-your-ai-assistant-obeys-it) (Claude, 6 min) ## Frequently asked questions **Which is worse, a false positive or a false negative?** It depends on the task. For payment scams a miss is usually worse. For an auto-reply to customers, a wrong action is usually worse. **Can I get rid of both?** Not completely. You can reduce them with clearer instructions and better examples, then catch the rest with a human step. **How does this show up in email sorting?** In the Gmail AI triage lesson, a newsletter labeled Lead would be a false positive, and a real lead left unlabeled would be a false negative. [All AI glossary terms, A to Z](https://learn.ynteractive.com/content/glossary) · [Free prompt templates](https://learn.ynteractive.com/content/prompts)