Skip to the stories
AI & ML

Deepfake Detection: Verify the Source, Context and Your Reaction

  • James Garcia
  • 308 reads

Visual clues are fading as AI improves. Check where a clip came from, seek corroboration and treat detector results as a reason to investigate.

Deepfake Detection: Verify the Source, Context and Your Reaction
A colorful collage of symbols and ripped paper forming the profile of a human head.

As generated images and video get harder to distinguish from authentic footage, the useful question is changing: instead of asking whether a face or hand looks wrong, ask who published the material, what context surrounds it and what it’s prompting you to feel or do.

That shift matters beyond misleading posts. Deepfakes can support scams and manipulation, and digital scams, including those involving deepfakes, caused more than $148 billion in losses in 2025, a rise of over 25 percent from the prior year. Yet spotting visual glitches is becoming less dependable. Marijus Briedis, chief technology officer at online security company NordVPN, says familiar flaws such as extra fingers and strange textures have mostly disappeared. Any new visual tell, he warns, may have a short life as image models improve.

Trace the post to its context

Cybersecurity experts increasingly recommend checking a post’s origins rather than examining pixels for signs of manipulation. A video may look convincing because the generation is good, because it has been edited, or because a short excerpt leaves out the context that would change its meaning. Visual inspection alone can’t resolve those possibilities.

Roei Sherman, senior director of research at cloud security firm Mitiga, cautions that publishing lists of visual detection tips can make them less useful: techniques described as clues can be incorporated into prompts. For a clip that appears to show a public figure, look for the full speech or interview and check whether other credible outlets have reported on the same event. A short clip that circulates without a clear origin deserves extra scrutiny.

Images can be harder to trace. A reverse image search may reveal earlier copies or related versions, but search results aren’t proof of authenticity. Sofia Rubinson, senior editor at misinformation watchdog NewsGuard, says AI-powered search can misattribute images by repeating the description supplied by people spreading a false claim. Treat search as a way to find leads, then check those leads against independent material.

The account that first posted a clip also needs scrutiny. Briedis suggests checking whether it has a history that goes back beyond a few weeks, whether the claim appears elsewhere with credible people or organizations attached, and whether the post is pressuring readers to act immediately. Those checks take attention, but they rely on context and behavior rather than visual patterns that can quickly become outdated.

Test plausibility and corroboration

Before sharing a striking claim, pause to consider whether the event makes sense. Would the person shown plausibly say or do this in public? Is there corroboration from multiple sources, or does the claim depend on one isolated clip? Plausibility can’t establish that something is false, but an implausible claim is a reason to look for more evidence.

Corroboration should mean more than seeing the same clip reposted across several accounts. Reposts may all lead back to one original upload, so they don’t provide independent confirmation. Look for accounts with their own reporting, footage from different angles or a longer recording that shows what happened before and after the excerpt.

This extra check is useful even for claims that seem believable. Social feeds mix real events with fabricated or distorted material, and an outrageous statement may turn out to be genuine while a plausible-looking image may not. The practical test is whether independent evidence supports the specific claim attached to the media.

Notice what the post is making you feel

Emotion is another clue—not to whether a video was generated, but to whether it may be trying to rush your judgment. A post that makes you furious, frightened or desperate to respond can encourage sharing before verification. Repetition can add to that pressure: if one dramatic clip keeps appearing, the volume of posts may create a sense of confirmation even when they all trace back to the same material.

Roei Sherman advises paying attention to that reaction. If a post seems designed to provoke an immediate response, take a moment before commenting or forwarding it. Check its origin and context first. That pause is especially useful when a post asks you to spread the claim quickly.

Use detection tools as a prompt to investigate

Specialist monitors and detection services can help surface known fakes or give you another reason to question a post. NewsGuard maintains an AI monitor; McAfee tracks AI-powered scams on its AI news blog; and AI or Not has a searchable database of known deepfakes and manipulated images.

These resources have limits. A detector’s result isn’t a definitive verdict, and a database can only help with material it has identified. Tools are most useful when something already seems questionable: they can provide another lead to check, but they don’t replace tracing the source, finding context and seeking independent corroboration.

As content authentication systems and detection methods develop, the reliability of any one approach may change. For now, the dependable habit is to treat detection results as a starting point. If you can’t establish where a striking image or clip came from and find supporting evidence, don’t share it. That small delay can keep an uncertain claim from gaining reach.

Source: James Garcia · mashable.com

From the same set