When AI fabricates citations, who checks before the court does?

The receipts are fake

Thomson Reuters has put out a warning that cuts to the heart of legal work: AI can fabricate citations that look real ^[1]. Not just wrong references or hallucinations in the conversational sense, but entirely invented case names, court names, dates and even plausible-sounding reasoning. The same goes for audio, video and documents.

The problem isn't that AI makes mistakes. The problem is that the fakes look convincing enough to slip past the people whose job is to catch errors.

How courts assume documents are real

Courts have always worked on a simple assumption: a document you submit is genuine until someone says otherwise ^[2]. If you hand in a contract, a witness statement or a legal memo, the court treats it as real unless the other side challenges it. That assumption is efficient. It keeps cases moving.

AI-generated fakes break that assumption. A fabricated citation doesn't just misstate the law; it invents a case that never existed. The judge, the opposing counsel and the client may all assume it's real, because it looks like every other citation in the bundle.

It's like turning up to a meeting with a fake bank statement. The numbers look right, the logo's in the right place, and nobody checks the account number until it's too late.

Lawyers are now the checkpoint

Thomson Reuters is clear: legal professionals must verify AI-generated content before it reaches a court or a client ^[1]. That means checking every case cited, every document referenced and every fact stated. The duty falls on the lawyer, not the AI.

Several U.S. courts have already sanctioned lawyers for submitting AI-generated fabricated citations ^[3]. The sanctions weren't light: fines, professional reprimands and public orders detailing what went wrong. The courts have made it plain that relying on AI without checking is not an excuse.

In the EU, transparency rules are tighter. Since 2 August 2026, people must as a rule be told when they are interacting with an AI ^[4]. That applies when AI drafts legal advice or generates documents for clients. The question becomes: if you're required to tell the client the AI wrote it, are you also required to tell them you checked it?

What this means for you

If you use AI to draft, research or summarise, here are three questions worth asking now:

• Who in your firm checks AI-generated work before it goes out, and do they know what to look for?
• If a citation turns out to be fake six months later, could you show what the AI gave you and when you spotted the error?
• Do your retainer terms or engagement letters say anything about AI use, and if they do, do they match what you're actually doing?

The risk isn't just professional embarrassment. It's sanctions, negligence claims and a loss of trust that's hard to measure and harder to rebuild.

How a digital witness fits

Here, the thing to seal is each search result, draft or summary the AI generates. You send a fingerprint of the AI's output to WitniumChain through the REST API, the SDK or the MCP server, and get a witness ID back. The content stays with you.

If a citation is later questioned, you can show exactly what the AI gave you, unchanged since that moment. Anyone you share the record with—your client, the court, your insurer—can check the fingerprint and see that it matches.

A digital witness doesn't prove the citation was correct. It shows what you were working with, and when you checked it. That's the difference between "I don't remember" and "here's the record".

The first 2,000 witnesses a month are free ^[5].

How it would work

1. The AI generates a memo citing three cases.
2. You seal the memo and get a witness ID.
3. You check the cases and find one is fabricated, so you correct the memo and seal the corrected version.
4. If it's challenged later, you can show both: what the AI gave you, and what you filed after checking.

What to watch

The transparency duties in Article 50 of the AI Act apply now ^[4]. Courts in several EU states are starting to issue practice directions on AI use in litigation. Keep an eye on your jurisdiction's rules, and on your professional body's guidance.

If you work with AI in legal research or drafting, the checkpoint is you. The question is whether you can show, if asked, that you did the checking.