False positives: checking a security alert
Learn what a security false positive means, how to assess a suspected detection error, and why restoring files or adding exclusions requires stronger evidence.
A security false positive is an incorrect detection that identifies a benign item or activity as malicious or as belonging to a threat category it does not meet.
Also called: false alarm, false detection, antivirus false positive, false positivesSuspected and confirmed errors are different
A false positive is a detection error. A familiar application being flagged, a clean result from another scanner, or a file that seems to work normally can raise the possibility of an error. None confirms it by itself.
For example, a small software publisher might release an update that a security product blocks. There are at least two questions to resolve: is this the authentic update, and did the detecting product assess it incorrectly? Trust in the publisher does not establish the integrity of the downloaded copy.
Microsoft’s false-positive guidance defines incorrect malicious detections and describes an investigation process. Its detailed portal procedures apply to managed Defender for Endpoint environments; they should not be treated as consumer Windows instructions.
Until the evidence settles the question, describe the alert as a suspected false positive. Keeping the item isolated gives you time to investigate without deciding that the alert is harmless.
False positive, false negative, and accurate detection
The result depends on whether the classification matches the item or activity. “Positive” means detected, not good.
| Outcome | What the security product reports | What is actually the case |
|---|---|---|
| True positive | A threat or relevant unwanted behavior is detected | The finding is accurate |
| False positive | A threat or unwanted category is detected | The item does not meet that category |
| True negative | No relevant threat is detected | The item is benign within the assessment |
| False negative | No relevant threat is detected | A threat was missed |
These terms describe an assessment, not a guarantee about an entire device. One benign file can be incorrectly flagged while a different malicious component is correctly detected.
An accurate PUP or PUA finding is also not necessarily a malware finding. Wanting to keep an application does not make its unwanted-software classification incorrect. Read the detecting vendor’s criteria before deciding whether the disagreement concerns facts or preferences.
Why an alert might be mistaken
Security products evaluate patterns, characteristics, behavior, and other signals. Some legitimate programs perform actions that resemble malicious activity: an administration tool may interact with other processes, for example. Similarity can be a reason for closer assessment without proving harmful intent.
The signature, heuristic, and behavioral-detection articles explain those methods. A method’s name does not establish whether a particular finding is correct.
There is also a difference between a threat detection and a reputation warning. A new or unfamiliar file may lack an established reputation. That is a different claim from identifying specific malicious code. Check the wording and detection source rather than treating every warning as interchangeable.
The reverse matters too. A malicious copy can borrow a familiar name, and an authentic-looking program can contain an unwanted component. Familiarity, normal operation, or a professional website are useful context, but they do not override an exact detection.
Gather evidence about the exact item
Before changing the security response, record what was flagged:
- The detecting product, full detection name, time, and completed action.
- The affected path or address, including whether the item was inside an archive.
- The application’s version and the actual publisher or download source.
- Relevant changes or behavior, including whether the item ran.
Protection History is the Windows event record; other products provide their own history. Compare the same file and version when seeking a second assessment. A report about an older release or a different download is not evidence about your copy.
A file hash can identify the exact bytes being discussed if the vendor asks for it. It does not certify that the file is safe. You do not need to run a flagged program to collect its displayed detection details.
Use vendor review and interpret second opinions
The detecting vendor is the appropriate place to request correction of its rule. Microsoft’s submission portal accepts reports of incorrect malware and PUA detections. Malwarebytes describes its own reporting route. Use the route for the product that raised the finding and preserve the response for the exact item.
A multi-engine service can add context, but its result is not a majority-vote safety certificate. The reasoning is simple: several “not detected” results still do not establish that a particular threat was absent. Different labels can also refer to different categories.
A local second-opinion scanner also needs comparable scope and exact-item evidence. If the first tool already quarantined the file, the second tool may be inspecting a different accessible set of objects.
VirusTotal explains that it aggregates multiple kinds of analysis and that standard submitted-file contents may be shared with partners and qualified customers. Do not upload private documents, credentials, or work files to a public analysis service without authorization. A workplace security team can choose an appropriate review process.
Why restoring or excluding first is a poor test
Restoring a quarantined item makes it accessible again. Excluding it changes what protection handles. Neither action demonstrates that the detection was wrong.
A broad folder exclusion can also affect future files unrelated to the disputed item. Microsoft explicitly notes that exclusions reduce protection. If an exception is necessary in a managed environment, its owner should assess the evidence and control its scope rather than treating it as a routine way to silence alerts.
Leave a suspected item isolated while gathering evidence. If the vendor confirms a correction, update the security product and reassess the affected item using its supported process. Restoration is a separate decision about a specific file, not an automatic consequence of finding an old online report.
Close the question with a documented outcome
A useful resolution records which detection was wrong, which file or activity was assessed, and what changed. A vendor correction may address the alert without proving that all copies or versions are benign.
If the evidence remains inconclusive, keep that uncertainty visible. Use an unaffected alternative or seek qualified assistance instead of forcing the file to run. For a managed device, let IT decide the response. The aim is an accurate assessment and a usable system, with protection changes tied to evidence.
Sources & documentation
Sources checked on 6 Oct 2026. Documentation research does not establish independent testing.