Deepfakes, edited screenshots, synthetic audio — the AI era makes it easier than ever to fake “proof.” If you care about truth, you can’t just ask, “What does this show?” You have to ask, “Where did this come from, and who touched it on the way here?” That trail is the digital chain of custody, and it’s the difference between real evidence and pretty fiction.
What changes when AI can fake anything
Before AI tools went mainstream, most people assumed photos and videos were honest unless proven otherwise. Now it’s flipped. A convincing clip can be generated on a laptop, shared in a group chat, and reposted a thousand times before anyone even asks if it’s real.
That doesn’t mean you can’t trust anything. It means you can’t trust anything without a trail. In the AI era, the chain of custody is not a bonus detail. It’s the whole ballgame.
The three layers of digital custody
Think of digital chain of custody in three layers:
- Origin: Where the file or claim first appeared. Which device, account, server, or institution.
- Integrity: Whether the file has been altered. Same hash, same metadata, same content.
- Transfer: Who handled it and how it moved — downloads, edits, reposts, archives.
You don’t always need lab-level forensics. But the more serious the claim, the tighter that chain needs to be.
Locking down origin in the AI era
For anything serious — elections, crimes, public safety — “I saw it on social media” is worthless. Start asking:
- Is this hosted on an official channel? Government site, court portal, known news outlet, or verified organization.
- Is there a stable URL? Not just a repost, but a page that can be revisited and archived.
- Does the publisher have skin in the game? Real organizations correct errors and can be held accountable.
AI can imitate faces and voices. It can’t fake a court docket number on an official server.
Checking integrity: hashes, metadata, and duplicates
This is the nerdy part, but it’s where the protection lives.
- Hash the file: Tools that create SHA256 or MD5 hashes give you a digital fingerprint. Any change to the file changes the hash.
- Save the metadata: Original creation date, device, and software info all matter. Stripped or scrambled metadata is a yellow flag.
- Compare versions: If three copies of a video exist, which one came first? Which matches the original hash?
You don’t have to do this for every meme you see. But if you’re documenting serious misconduct or building a public case, this stuff matters more than the caption.
Documenting transfer: who touched the file
In a world where AI can tweak footage with one click, every handoff is a risk. That’s why investigators track who collected, copied, edited, and shared a file.
- Keep a simple log: date, time, person, and action (“downloaded from X,” “copied to secure drive,” “shared with reporter”).
- Limit edits: work from copies, keep the original locked away.
- Don’t pass files through apps that compress or “enhance” them unless you keep the original intact.
If the story ever goes to court, a sloppy chain of custody lets bad actors say, “This could have been altered at any point.” Don’t give them that opening.
AI as a tool for truth, not just lies
The same technology that makes fakes can also help confirm the real thing. There are tools that detect manipulation, compare faces, and spot inconsistent lighting or shadows. Some newsrooms and NGOs already use AI to flag suspicious media before publication.
The key difference: who is using the tech and what they publish with it. A responsible outlet explains how it verified something. A bad actor just says, “Trust me, bro.”
What regular people can actually do
You don’t need a lab. You do need better habits.
- Save links and original files when you see something important, not just reposts.
- Archive pages using tools that timestamp content so it can’t be quietly edited later.
- Ask for sources in public when big claims are made. Make “Where did this file come from?” a normal question.
- Refuse to share untraceable content no matter how good it feels to believe it.
Why this matters more than ever
Powerful people already understand the value of muddying the record. AI just handed them a better fog machine. If every clip can be called “fake” and every fake can be sold as “real,” the only thing left to stand on is documented chain of custody.
That’s the whole point of this project: you don’t have to trust anyone’s personality. You can trust the trail.
Keep reading next
For a hands-on guide to using these habits in your own work, read next: The Evidence Matters Toolkit: Verify, Prove, and Share the Truth.