Can You Prove a Video Isn’t a Deepfake?

Can You Prove a Video Isn’t a Deepfake? What a Timestamp Can and Can’t Show

A timestamp can’t prove a video is genuine. What it can do is prove that one exact file existed no later than a specific moment, and that turns out to be useful in more deepfake disputes than you’d expect. It won’t tell anyone who recorded the video, whether the scene was real, or whether the pixels were generated. It does fix the video in time, so nobody can later swap in a different file and claim it was the original.

This article walks through where that narrow guarantee helps, where it doesn’t, and how to set up a routine so the record is worth something if you ever need it.

A fingerprint with a date on it

Start with what’s actually being stamped. A cryptographic hash function like SHA-256 reads every byte of a file and produces a 64-character fingerprint. Change one frame, one audio sample, or one byte of metadata, and the fingerprint comes out completely different. Same bytes, same fingerprint, every time.

A timestamp ties that fingerprint to a moment. If the fingerprint is written into a public blockchain, anyone can later see that it was recorded at a given time, and nobody has to take Certelo’s word for it. If you can then produce a file that hashes to the same value, you’ve shown that this file, byte for byte, existed by then.

Notice what’s missing from that sentence. Nothing about the content. Nothing about who made it. The timestamp is deliberately blind to what the video shows.

Why spotting fakes isn’t the whole answer

Most advice about deepfakes is about detection: look at the blinking, the teeth, the lighting on the cheekbones. Detection tools help, but they play a game that keeps moving. They try to identify what’s fake, and a generator that improves tomorrow can beat the detector that worked today.

There’s also a quieter problem. Detection can’t help the person whose real video gets dismissed. Once convincing fakes are common, anyone caught on camera can say “that’s AI,” and a genuine recording has to defend itself. Legal scholars call this the liar’s dividend. The accused doesn’t need to prove the video is fake, only to make people unsure.

That’s where a record made before the argument started becomes valuable. It can’t prove the video is true, but it can take a few easy lies off the table.

Three situations where an early timestamp earns its keep

These are hypothetical examples to show the logic, not real cases.

The backdated “original.”

A local reporter films a street incident on her phone and anchors the file’s hash that evening. Three weeks later, a polished version appears online, claiming to be the true uncut footage and saying hers was edited. Hers has a public record from the night of the incident. The other one doesn’t. That doesn’t make hers authentic by itself, but it shows hers existed first, and anyone claiming theirs is the source now has to explain the gap.

The “that’s AI” defense.

A warehouse manager records a safety violation. Later, the contractor says the clip was generated. If the manager hashed the file the day he recorded it and kept the original, he can show the exact file he’s presenting is the one that existed then, and that nothing has been regenerated or swapped since. The contractor’s claim now has to be about the original capture, not about a file that might have been touched later.

Which version came first.

Two editors argue over whose cut of a documentary came first, and whether one was derived from the other. A timestamp for each export gives a verifiable order. It doesn’t settle who made what, but it settles which file existed when.

Where a timestamp does nothing

This is the part most vendors skip.

A deepfake can be timestamped too.

If someone generates a fake and anchors it the same minute, the fake gets an equally valid timestamp. A timestamp tells you about time, not truth. It’s honest, and that’s both its value and its limit.

It can’t see before the stamp.

If you hash a video after you’ve edited it, the record covers the edited file. A hash made today says nothing about whether the clip was manipulated yesterday.

Garbage in, garbage out.

If you stamp a file that was already fake when you got it, you’ve perfectly preserved a fake.

The bytes have to match.

Upload a video to a social platform and it gets recompressed. The copy people see no longer hashes to your fingerprint, even though it looks identical. Verification works against the exact file you kept, not against whatever circulates online.

It doesn’t identify anyone.

A blockchain record doesn’t say who recorded the file or who stamped it.

Put together: a timestamp is good at defeating a claim that a file is newer than it is, or that a different file is the original. It’s poor at answering whether the footage is true.

Where C2PA fits, and where it doesn’t

The other approach getting attention is C2PA, whose signed “Content Credentials” travel with the file and record where it came from and what was done to it. It’s built for the opposite job: capture-time provenance from a supporting device or tool.

It has well-known limits, and the people discussing it are fairly open about them. One broadcast-industry analysis points out that C2PA does not independently prove that each assertion is accurate, that nothing happened before the first signature, or that the signer deserves trust (TV Technology). Credentials can also disappear: Scientific American notes that adoption has been slow and the credentials often get stripped as files circulate online. And a missing credential doesn’t make a video fake, only unverifiable through that route (Magic Hour’s practical guide makes the same point).

So the two aren’t rivals. C2PA, where a device supports it, tries to say something about the capture. A separate timestamp, made independently, adds a record that doesn’t depend on the credentials surviving the journey. If a platform strips the metadata, a hash you recorded earlier is still sitting on a public ledger. Different failure modes, which is the reason to use both.

A capture-day routine that makes the record worth something

The record is only as good as what you do around it. A routine that holds up:

1- Keep the original. Don’t edit, trim, or “clean up” the first file. Copy it and work on the copy.

2- Hash it quickly. The shorter the gap between recording and stamping, the less room for anyone to argue about what happened in between.

3- Stamp the companions too. Raw audio, a second angle, screenshots of the device’s own file details, photos of the scene. A fake has to be consistent with all of them.

4- Stamp each export separately. Every version you publish gets its own record, which builds a visible chain from raw capture to released cut.

5- Write down the context. Date, place, device, who was present. This is your own notes, not blockchain evidence, but it’s what makes the hash meaningful later.

6- Store the files safely. A record you can’t match to a file is useless. Keep originals somewhere durable, with backups.

7- Never overwrite. If a file changes, it’s a new file with a new fingerprint.

What Certelo does at the hashing step

With Certelo, the SHA-256 hash is calculated locally, on your own device. The video isn’t uploaded just to create the proof. For anyone handling sensitive footage, such as unreleased interviews, internal security video, or material under NDA, that’s a practical difference, since the proof can exist without the footage leaving your hands.

The hash is then anchored to the Electra Protocol blockchain, which provides a public, independently checkable record that the fingerprint existed at or before the recorded time.

Two clarifications, because the terminology gets muddled. “Trusted timestamping” in its traditional sense usually means a token issued by a time-stamping authority under a standard like RFC 3161 (our RFC 3161 explainer covers it). Blockchain anchoring is a different trust model: you rely on the public ledger rather than a single authority. Certelo’s record isn’t a qualified electronic timestamp under EU rules, and this article makes no claim that it satisfies any particular legal requirement.

Checking a record later

Verification is the whole point, so it’s worth knowing how it works. You take the original file, calculate its hash again, and compare it to the hash recorded on the blockchain. Matching hashes mean the file is identical to the one that was stamped. If you only have a certificate and hash information, you can still check the record against the ledger, but only the original file can show that a specific video matches it.

If someone hands you a video claiming it was stamped, the right question isn’t “is there a timestamp?” It’s “does this exact file produce the recorded hash?” If it was re-encoded, trimmed, or compressed, it won’t.

How courts are handling deepfake claims

Be careful with legal generalizations here. This section is about US federal courts only. It isn’t legal advice, and other jurisdictions work differently.

Under US Federal Rule of Evidence 901, the person offering evidence must produce enough to support a finding that it is what they say it is. Deepfakes have prompted discussion about whether the federal rules need changing. As of the reports I found, the picture is unsettled: in May 2026 the Advisory Committee on Evidence Rules declined to advance two draft AI-related rules, and it also declined to publish the proposed Rule 901(c) for public comment. The draft text was prepared but not approved. The draft’s approach was a two-step process: if the challenger offers enough evidence to support a finding of a deepfake, the proponent would then have to show the item is more likely than not authentic. The committee has also acknowledged that the rule doesn’t address deepfakes making juries distrust genuine evidence. A deepfake agenda item was again on the committee’s October 2026 meeting list, so check current status before relying on any of this.

For the practical point: if authenticity is ever challenged, a clear chain (original file, early hash, documented handling) is the kind of supporting material that can help a proponent show what a file is. It doesn’t guarantee admissibility. For more on that side, see our articles on self-authenticating evidence and the legal validity of blockchain records.

Summary

A timestamp won’t prove a video is real, and anyone who says otherwise is overselling. It will show that a specific file existed by a specific time and hasn’t changed since, which is enough to answer some of the most common disputes: backdating, swapped files, and “it was made after the fact.”

If you’re going to rely on it, do the boring parts well. Keep originals, hash early, stamp each version, and document the context. When you’re ready, you can create a timestamp record with Certelo and verify it independently later.

Let's clarify more

Can a timestamp prove a video isn’t AI-generated?

No. A timestamp shows when a file existed, not how it was made. It can help rebut claims that a file was created later than it was, or that it’s been altered since stamping.

Can deepfakes be timestamped too?

Yes. Anyone can hash and timestamp any file, including a fake. The timestamp is neutral about content, which is why it should be paired with other evidence.

If my video is re-uploaded to social media, does the proof still work?

The record stays on the blockchain, but platforms usually recompress video, so the circulating copy won’t hash to the same value. Verification works against the exact original you kept.

Is C2PA better than a blockchain timestamp?

They do different things. C2PA tries to attach signed provenance information at capture or editing, while a timestamp independently records that a file existed at a given time. Metadata can be stripped from files, whereas a hash already anchored on-chain isn’t affected.

Does Certelo upload my video?

The SHA-256 hash is calculated locally on your device, and the original file isn’t uploaded merely to create the proof.

Does a Certelo record prove I made the video?

No. It provides evidence that a particular digital fingerprint existed at or before the recorded time. It doesn’t establish authorship, ownership, or identity.

Will a court accept a blockchain timestamp as evidence?

That depends on the jurisdiction and the case. It can contribute to an evidentiary record, but admissibility is decided by the court, and this article isn’t legal advice.

What happens if the video is edited after I stamp it?

Any change, even re-saving with different settings, produces a different hash. The stamp still covers the original, but the edited file needs its own record.