The Liar’s Dividend: Why Deepfakes Can Make Real Evidence Easier to Deny
The obvious danger of deepfakes is that people may believe something false. A fabricated video can create a fictional event, a cloned voice can impersonate a relative, and a generated image can be mistaken for a photograph. There is another danger that receives less attention: people may stop believing genuine evidence.
This effect is often called the liar’s dividend. When society knows that realistic fabrications are possible, a person caught in an authentic recording can claim that the recording is artificial. The claim does not need to convince everyone. It only needs to create enough doubt among supporters, audiences, or decision-makers to delay accountability.
The result is a new information problem. We must improve our ability to identify false media without allowing the existence of deepfakes to turn all evidence into “just another possibility.”
How the liar’s dividend works
Imagine that a real video shows a public figure making a damaging statement. The video is challenged immediately. Supporters say it was generated or edited. The public debate shifts away from what was said and toward whether anyone can prove the file’s authenticity. While analysts investigate, the original claim may spread, disappear, or become trapped in partisan disagreement.
The strategy is effective because it exploits uncertainty. Most viewers do not have the original file, the recording equipment, or the technical expertise needed to examine it. They must rely on institutions, journalists, platforms, or experts. If those institutions are already distrusted, a simple denial can be enough to prevent consensus.
The dividend is especially powerful when the person making the denial benefits from delay. A correction may arrive after the audience has moved to another story. Even if the recording is later authenticated, some people will remember only that experts disagreed.
Why detection alone is not enough
A detector can estimate whether a file has characteristics associated with synthetic generation. It cannot answer every question about context, editing, or the event itself. A real recording may have been trimmed. A genuine voice may have been enhanced. A file may have been re-encoded by a social platform. These processes can produce technical signals that resemble manipulation.
Conversely, a sophisticated fake may avoid the signals a detector expects. Systems are trained on particular examples and can perform poorly on new generation methods. A result from one tool should therefore be treated as evidence to weigh, not a verdict that ends the inquiry.
A stronger process examines the whole chain of evidence. Where did the file originate? Who had access to the recording? Are there independent recordings from different angles? Does the background match the location and date? Is there a longer version? Do witnesses, transcripts, or contemporaneous reporting support the event?
Provenance and chain of custody
In journalism, law, and scientific work, chain of custody describes how evidence was collected, stored, transferred, and analyzed. Digital media can benefit from a similar approach. The original file should be preserved when possible. Its creation time, source, hash, edits, and publication history should be recorded. Analysts should document which copies they examined.
Content provenance standards such as C2PA can support signed information about a file’s origin and editing history. Provenance is not a magic authenticity stamp. It can be incomplete, absent, or attached after the fact. However, a transparent record is stronger than an anonymous repost or screenshot.
Newsrooms and institutions should communicate the limits of their findings. Instead of saying “this video is definitely real” without explanation, they can describe the evidence: the file was obtained from the original device, matches independent recordings, contains a continuous timeline, and has no detected alteration in the examined regions. Clear reasoning makes a conclusion easier to evaluate.
The role of trusted institutions
Deepfake controversies often become contests between competing authorities. One expert says the file is fake; another says it is real. Audiences may choose the conclusion that matches their beliefs. Institutions can reduce this problem by publishing methods, preserving source material, and correcting errors visibly.
Journalists should identify what they know and what remains uncertain. A responsible report can state that a video is authentic as a file while noting that the caption may be misleading. It can explain whether the analysis covers the entire recording or only a viral excerpt. It can distinguish technical authenticity from the truth of a person’s claims.
Public organizations should prepare before a crisis. They need secure publishing channels, media verification contacts, and a process for responding to impersonation. Waiting until a fake appears makes it harder to establish which channel is genuine.
How individuals can reason under uncertainty
People do not need to make a binary decision immediately. It is reasonable to say, “This claim is unverified,” while looking for better evidence. The important distinction is between healthy uncertainty and automatic dismissal. The possibility that deepfakes exist does not make every recording equally doubtful.
Use proportional skepticism. A casual meme does not require the same investigation as evidence of criminal conduct or a public emergency. Ask who benefits from the denial, whether the person has a history of making unsupported claims, and whether independent evidence is available.
Avoid repeating a false claim unnecessarily while correcting it. When discussing a disputed recording, link to the best available evidence and explain the verification status. Do not make the fake more memorable than the correction.
Rebuilding trust without demanding perfection
No verification system will remove uncertainty from digital media. Cameras can be hacked, files can be edited, witnesses can be mistaken, and experts can disagree. The objective is not perfect certainty. It is a more reliable process for evaluating evidence than intuition, virality, or partisan loyalty.
That process includes source transparency, secure archives, provenance, independent corroboration, skilled analysis, and honest communication about limits. It also includes consequences for deliberate deception. Platforms, organizations, and publishers should distinguish clearly between labeled fiction, satire, misleading manipulation, and fraudulent impersonation.
Deepfakes challenge the old assumption that seeing is believing. They should not lead us to the opposite assumption that seeing proves nothing. The responsible position lies between those extremes: visual and audio evidence remain valuable, but their meaning depends on origin, context, corroboration, and a documented chain of trust.
References
[1] C2PA Technical Specification
