For a long time the fear was that fakes would become good enough to deceive. The subtler danger is the reverse: it is enough that fakes exist for the truth to become contestable. When everyone knows that an image, a voice or a video can be fabricated, the liar no longer needs to produce a fake: faced with damaging evidence, it is enough to say "it's a fake".
1 The liar's dividend
The concept, and its paradox.
The definition
The fake first benefits those who lie
The "liar's dividend" is a concept formulated by the legal scholars Robert Chesney and Danielle Citron in the California Law Review (2019). The idea: as the public learns that anything can be faked, those caught out can dismiss genuine evidence by declaring it fabricated. The paradox is cruel: the more people are educated about deepfakes, the more the liar's argument is reinforced. The dividend grows, they write, "perversely", in proportion to the success achieved in raising public awareness of fakes. The threat is not only the fake that circulates; it is the generalised suspicion that it licenses.
2 The effort asymmetry
Why accusing and proving do not weigh the same.
The imbalance
An accusation takes three words, a proof takes three weeks
The driver of the dividend is an asymmetry of cost. Accusing is free, instant and understood by all: "it's a fake" is said in a second. Proving authenticity is slow, costly and hard to read: it takes forensic expertise, tools, technical vocabulary, and the verdict always arrives after the doubt has spread. This asymmetry is not specific to deepfakes: it is a particular case of "Brandolini's law" (Alberto Brandolini, 2013), according to which "The amount of energy needed to refute bullshit is an order of magnitude bigger than that needed to produce it". The cause is also cognitive: a false but simple and appealing claim is adopted effortlessly, where a nuanced truth requires one to pause. Refuting will always cost more than asserting.
3 Digital provenance
The countermeasure: certify the real at the source rather than hunt the fake.
Signing the real
Attaching a proof of origin to the media
Since detecting every fake is a losing race, another path is to certify the real at the source: this is "digital provenance". It attaches to a media file, from the moment of its creation, a cryptographic certificate attesting where it comes from and how it has been modified. The open C2PA standard and its "Content Credentials" layer brought together more than 5,000 members in 2025; Google has watermarked more than 10 billion pieces of content with SynthID; cameras sign the image at the moment of capture (Leica, Sony, Samsung). The aim is not to prove that a file is fake, but to prove that a file is what it claims to be.
C2PA coalition (2025)
> 5,000
members adopting the "Content Credentials" standard.
SynthID watermarks
> 10 billion
pieces of content marked by Google's tool.
4 The limits
Provenance shifts the burden, it does not remove it.
What the signature does not say
Proving a signal is not proving the truth
Provenance is no silver bullet. Watermarks can be removed or imitated: researchers have shown as much. Above all, a signature attests to an origin, not to a truth: a genuinely captured image can show a deceptive scene. As Sam Gregory, of the NGO WITNESS, sums it up, provenance "provides signals but does not prove truth". And a trap lurks: if we grow used to demanding a certificate, the absence of a certificate will end up passing for proof of a fake, even though the vast majority of genuine content carries none. Finally, the dividend's effect remains debated: it is real but bounded, mediated by partisanship and tempered by verification.
5 Takeaways
To remember.
✓
The liar's dividend: when anything can be faked, claiming "it's a fake" is enough to dismiss genuine evidence (Chesney & Citron, 2019).
✓
An asymmetry of cost: doubting is free and instant, proving authenticity is slow and costly (Brandolini's law).
✓
Digital provenance (C2PA / Content Credentials, SynthID, signing at the moment of capture) certifies the origin of a media file.
✓
It shifts the burden without erasing it: a signal is not a truth, and the dividend's effect remains bounded and mediated by partisanship.
This notion sheds light on an analysis
First published: June 27, 2026