In May 2026, a federal judge in Oregon sanctioned two lawyers $110,000 — the largest AI hallucination penalty in U.S. legal history — after they submitted 23 fabricated citations and eight invented quotations to the court. It wasn't an isolated incident. A database tracking these cases now counts over 1,000 U.S. court decisions, and more than 1,490 worldwide, where a party relied on AI-generated material that turned out to be fake — including a Nebraska attorney suspended in April after 57 of 63 citations in one brief were defective, and a Mississippi trial cancelled in June after both sides cited cases AI had invented.
The pattern is always the same: someone asks a general-purpose AI to summarize or research something, it answers with total confidence, and nobody checks before filing. The answer sounds plausible. It just isn't real.
Why this is a contract-review problem too
Most of the attention on "AI hallucination" has focused on fake case law in litigation. But the same failure mode exists in any tool that uses AI to read and analyze contracts — extracting obligations, renewal deadlines, risk clauses, payment terms.
Industry estimates put the hallucination rate at 5–12% when general-purpose AI tools analyze contract types outside their training data — high enough that an invented renewal deadline, or a payment obligation that doesn't actually exist in the document, can slip through unnoticed until it becomes a real problem.
The difference between a useful legal AI and a risky one isn't the underlying model. It's whether you can verify what it told you before you act on it.
A checklist for evaluating any contract AI
Before you let an AI reading your contracts influence a business decision, check four things:
| Criterion | Question to ask | Why it matters |
|---|---|---|
| Traceability | Does the answer point to the exact spot in the document it came from? | Without this, you can't check whether the AI read it correctly — you can only trust it blindly |
| Human editing | Can you manually correct a wrong extraction? | No AI is right 100% of the time; the system needs to admit that and let you fix it |
| Re-analysis | Can you request a fresh scan when the contract changes? | Contracts get amended; a stale extraction is as dangerous as a wrong one |
| Uncertainty signaling | Does the AI flag when it's unsure, or does it always answer with the same confidence? | An AI that never hesitates is the one most likely to mislead you |
If a tool fails these four checks, the risk isn't hypothetical — it's a matter of time before a mis-extracted obligation turns into a missed deadline, or worse, a business decision made on bad information.
How DottSign was built around this
When we shipped automatic obligation extraction, the starting question wasn't "which model answers fastest" — it was "how do we make sure the answer can actually be trusted."
In practice:
- Every citation is a clickable card that jumps straight to the exact spot in the source PDF. You don't have to take the AI's word for it — you verify the original in one click.
- Every extracted obligation can be edited manually. If the AI got something wrong or read a clause differently than you intended, the fix is yours to make, not a support ticket to file.
- A fresh analysis can be requested anytime — useful whenever a contract is amended or renegotiated.
- The AI assistant cites its source on every answer, with links that jump straight to the right spot in the document.
That's the difference between an AI that sounds trustworthy and one that is verifiable. The first is what's generating six-figure sanctions in courtrooms right now. The second is what you can actually build business decisions on.
The bar is only going up
With over 1,000 U.S. court decisions already dealing with AI-fabricated citations — and cases growing month over month — the direction is clear: courts, clients, and regulators are going to keep demanding that AI tools show their work, not just their conclusions.
Choosing a contract management tool today is also choosing which side of that line your business will be on — the side that can prove where every piece of information came from, or the side that finds out the hard way.
See it for yourself
Create a free DottSign account and see how obligation extraction with traceable citations works on your own contract — free to start.