We tend to believe that seeing is knowing. In reality, this is a very recent habit, born with photography in the 19th century and lasting only about a hundred and fifty years. Before it, no one granted evidentiary value to an image: a painting showed what the painter chose to show. This parenthesis may be closing.
The California law that came into force this weekend, whose labelling requirements we detailed, is a technical response to a problem that is first and foremost philosophical. Let us take the time to examine it.
Why photography worked as evidence
The evidentiary force of photography did not come from its fidelity to reality, which has always been debatable. It came from an economy. Faking an image required equipment, time, and rare expertise. The cost of the fake was so high that one could reasonably presume authenticity by default.
Faking has always existed, from the early days of photography, and entire regimes erased figures from official photographs. But that remained the business of states and specialists. The presumption of authenticity held statistically.
That economy has reversed. Producing a convincing image no longer costs anything, as shown by the models we cover, from Seedance to FLUX 3, which now generate sound-laden videos dozens of seconds long. When the fake becomes free, the presumption of authenticity loses its foundation.
The real risk is not the one we think
We usually imagine the danger as a fake so convincing it would fool everyone. That is a real risk, but probably not the main one.
The most corrosive risk is the opposite: when anything can be fake, anything can be denied. An authentic and inconvenient piece of evidence becomes contestable with a mere shrug. It is no longer necessary to prove a document is doctored; it is enough to suggest it might be. Researchers call this the liar's dividend: the benefit that someone with something to hide derives from general uncertainty.
The consequence is not that we will believe anything. It is that we risk believing nothing at all, which amounts to believing whatever suits us. This is a concern we touched on from another angle in our article on deepfakes.
If the image alone no longer proves anything, what does? The answer emerging is the chain: who produced this content, with what tool, when, and who can vouch for it. That is exactly what the C2PA standard builds, which we explain in a dedicated article. We no longer trust the image; we trust its traceability and the one who signs it.
The return of something old
This shift feels like déjà vu, and that is rather reassuring. Before photography, information already circulated, and people knew how to assess it. They did not ask whether an image was authentic; they asked who is telling this, what do they know, what interest do they have in telling me this.
That is exactly the mode of verification we are returning to. Trust moves from the content to the source, and from the content to corroboration: a real event leaves multiple traces, with multiple witnesses, from multiple angles. A fake, even an excellent one, is generally alone.
These reflexes are those we detailed in our verification method: cross-check, trace back to the source, and be wary of your own satisfaction when information suits you.
What this demands of us
There is a new requirement, and it is uncomfortable. For a century and a half, visual trust was free: you just had to look. It is becoming costly again: it requires checking a source, cross-referencing, and suspending judgment.
Yet we consume images at a pace that makes such vigilance impossible to sustain at all times. No one is going to verify the provenance of every photo encountered in a feed. Realism therefore lies in calibrating one's effort: accepting not to know for the trivial, and demanding more when the stakes matter—that is, when an image is used to accuse, to decide, or to persuade.
There remains a note of optimism. What we are losing is a convenient shortcut that worked for a historically brief period. What we are regaining is a skill humanity practised for millennia: assessing testimony rather than swallowing it whole. This is not a regression. It is simply more tiring, and probably more honest.