A hallucination is a confident statement the sources do not support. In legal due diligence there are two kinds: invented findings and missed findings. You reduce both by grounding every statement in the documents, controlling what the AI is allowed to write, and keeping lawyers in charge of the result.
What is an AI hallucination?
An AI hallucination is output that sounds right but is not supported by the source. Examples in a legal review:
- A clause that does not exist
- A wrong notice period or threshold
- A citation to the wrong document or paragraph
- An obligation read into a contract that is not there
Why does it matter in legal due diligence?
Findings drive price, warranties, indemnities and sometimes the decision to sign. Two errors matter:
- False positives: invented issues waste time and damage trust.
- False negatives: a missed change-of-control clause or guarantee can cost real money.
The second is often more dangerous, because nobody notices it.
What does "hallucination-resistant by design" mean?
It means the system is built to make errors unlikely and visible. It does not mean the system is error-free. Typical design principles:
- Grounding: every statement must be backed by a passage in the data room.
- Controlled output: the AI works within a defined review method instead of writing freely.
- Full reads: a clause is only reported as missing after the whole document has been read.
- Checks before delivery: findings are checked before they reach the user.
- Visible uncertainty: points that could not be confirmed are flagged, not hidden.
Why does traceability matter?
Traceability turns trust into something you can test. If every finding links to its source paragraph, a lawyer can check it in seconds.
Why do lawyers stay in charge?
Because no system is perfect, and because legal advice requires judgment. AI drafts. Lawyers decide.
FAQ
Can hallucinations be eliminated completely?
No. They can be made rare and easy to spot.
Is a bigger AI model the answer?
Not on its own. How the system is built around the model matters more.
How do I test a tool for hallucinations?
Run it on a deal you know well. Check every citation and look for what it missed.
- ✓ Hallucinations include invented findings and missed findings.
- ✓ Ground every statement in the source and make uncertainty visible.
- ✓ Traceability lets lawyers verify findings quickly.
- ✓ AI drafts. Lawyers decide.
Fusewise is an AI legal due diligence platform for M&A deal teams, based in Zurich.
