When a fake looks real: *how courts decide* what counts as evidence

The problem arrives faster than the answer

A judge in Indiana recently dealt with video evidence that may have been altered by AI. ^[1] The case turned on whether what the recording showed had actually happened. The video looked fine. The technical analysis wasn't conclusive. In the end, the court had to weigh credibility the old-fashioned way.

That case won't be the last. AI tools that generate or alter photos, video and audio are cheap, fast and convincing. The tools that detect manipulation are slower, often require the original file and regularly produce uncertain results. Courts are being asked to decide what happened based on evidence that might have been invented last week.

Think of it like this: a receipt proves you bought something, until receipts become trivial to forge at home. At that point, the shop's own sealed record, timestamped and unchanged, becomes the one that holds up. The difference is that with digital content, we've reached that point already.

How courts assess evidence today

In Sweden, as in most of Europe, there is no hierarchy of proof. Judges assess evidence freely under the principle of fri bevisprövning. ^[2] Anything can be presented—a photo, a recording, a printout, a witness statement—and the court decides what weight to give it.

That flexibility is useful when facing something new, but it also means the burden falls on the parties. If you present a recording and the other side questions it, you need to show it is what you say it is. A file on its own proves nothing about when it was made or whether it has been edited.

The AI Act adds transparency duties but not evidential rules. From 2 August 2026, people must be told when they interact with an AI system, and AI-generated content must carry a machine-readable marking. ^[3] A deepfake—content that resembles real people or events and would be mistaken for authentic—must be disclosed by whoever deploys it. ^[4] Those duties matter, but they don't answer the question a court will ask: is this recording the one made on the day in question, or has someone changed it since?

What it takes to show a record hasn't changed

Authentication has always been part of presenting evidence, but it used to be simpler. A signed contract, a dated invoice, a letter with a postmark: the form itself carried some assurance. Digital files don't.

A video file has metadata, but metadata can be rewritten. A screenshot shows something, but screenshots can be faked in seconds. Cloud storage keeps versions, but the person with administrator access can delete them. If you are the one keeping the record, and you are also the one being questioned, your own testimony about the file's integrity isn't worth much.

The strongest position is to show that the record existed at a known time and that any change since then would be visible. That requires something outside your own control.

One American legal commentator put it plainly: in a dispute over AI-generated evidence, the side that can produce a verifiable chain of custody for their digital files will have an advantage over the side that cannot. ^[5] Verifiable means checkable by someone else, not just asserted by the person who benefits.

How a sealed record changes that conversation

This is where a digital witness fits. It seals a record at a known time, so that anyone can later check whether it has changed.

Here, the thing to seal is the file itself: a photo, a video, a document, an audio recording. Sealing takes a fingerprint of the file, not the file, so the content never leaves your systems. The fingerprint is a short string of characters derived from the file. Change one pixel or one word and the fingerprint changes completely.

The fingerprint is sent to WitniumChain, which writes it to a chain where any later tampering would show. You get a witness ID back. To check, you calculate the fingerprint again and compare it with the sealed one. If they match, the file is exactly what was sealed on that date. Anyone you share the record with can do the same check, including a court, an expert or the other party in a dispute.

What it shows: that the file hasn't changed since it was sealed. What it doesn't show: that the content is true, or that the file shows what you say it shows. A sealed fake is still a fake. But if the question is whether your record has been edited, a seal answers it.

How it would work

1. Capture the file: a photo on site, a contract sent to a client, a recording of a meeting.
2. Seal it immediately, through the REST API, the SDK or the MCP server. The fingerprint goes on the chain; the file stays with you.
3. When someone questions it, recalculate the fingerprint and compare. If it matches, nothing has changed.

What this means for you

If your organisation handles evidence that might one day be disputed—photos, contracts, recordings, decisions—three questions are worth asking now:

• When is the record created? The earlier you seal it, the harder it is for someone to argue it was made or edited later.
• Who keeps it, and who might question it? If you are both the keeper and a party with an interest, an independent seal makes your record harder to challenge.
• Could you show a court that the file hasn't changed? Metadata and cloud logs help, but they aren't proof. A sealed record is.

You don't need to seal everything. Seal what matters: the things that could be questioned and where being able to show "this is exactly what we had on that day" would settle an argument before it grows.

The first 2,000 seals a month are free. The documentation is public. Whether you build it into a case management system, a media workflow or a customer service platform, the idea is the same: make it easy to seal records as they are created, so that when one is challenged, the answer is short.

What to watch

The EU's AI Act transparency duties apply from 2 August 2026. That is the date by which AI-generated content must carry machine-readable marking, and deepfakes must be disclosed.

Beyond that, courts will continue to set the evidential standards case by case. The technical arms race between generation and detection will go on. What you can control is whether your own records can be shown unchanged, or whether you are left arguing about a file that might as well have been edited this morning.