'/%3e%3crect%20width='440'%20height='260'%20fill='url(%23gl5)'/%3e%3cg%20stroke='%236f86c4'%20stroke-width='1.4'%20opacity='0.22'%20fill='none'%3e%3cpath%20id='l5a'%20d='M120%2082%20L220%20182'/%3e%3cpath%20id='l5b'%20d='M220%2072%20L220%20182'/%3e%3cpath%20id='l5c'%20d='M320%2082%20L220%20182'/%3e%3c/g%3e%3cg%20fill='none'%20stroke='%237e92cf'%20stroke-width='2'%20opacity='0.4'%3e%3ccircle%20cx='120'%20cy='82'%20r='9'/%3e%3ccircle%20cx='220'%20cy='72'%20r='9'/%3e%3ccircle%20cx='320'%20cy='82'%20r='9'/%3e%3c/g%3e%3ccircle%20cx='220'%20cy='182'%20r='13'%20fill='none'%20stroke='%239fb2e8'%20stroke-width='2'%20opacity='0.5'/%3e%3ccircle%20cx='220'%20cy='182'%20r='3.5'%20fill='%235fe0d6'%20opacity='0.55'/%3e%3ccircle%20cx='220'%20cy='182'%20r='6'%20fill='none'%20stroke='%235fe0d6'%20stroke-width='1.6'%3e%3canimate%20attributeName='r'%20values='6;22'%20dur='3s'%20begin='1.4s'%20repeatCount='indefinite'/%3e%3canimate%20attributeName='opacity'%20values='0.5;0'%20dur='3s'%20begin='1.4s'%20repeatCount='indefinite'/%3e%3c/circle%3e%3ccircle%20r='3.2'%20fill='%235fe0d6'%3e%3canimateMotion%20dur='3s'%20begin='0s'%20repeatCount='indefinite'%3e%3cmpath%20xlink:href='%23l5a'/%3e%3c/animateMotion%3e%3canimate%20attributeName='opacity'%20values='0;0.9;0.9;0'%20keyTimes='0;0.1;0.85;1'%20dur='3s'%20begin='0s'%20repeatCount='indefinite'/%3e%3c/circle%3e%3ccircle%20r='3.2'%20fill='%235fe0d6'%3e%3canimateMotion%20dur='3s'%20begin='0.6s'%20repeatCount='indefinite'%3e%3cmpath%20xlink:href='%23l5b'/%3e%3c/animateMotion%3e%3canimate%20attributeName='opacity'%20values='0;0.9;0.9;0'%20keyTimes='0;0.1;0.85;1'%20dur='3s'%20begin='0.6s'%20repeatCount='indefinite'/%3e%3c/circle%3e%3ccircle%20r='3.2'%20fill='%235fe0d6'%3e%3canimateMotion%20dur='3s'%20begin='1.2s'%20repeatCount='indefinite'%3e%3cmpath%20xlink:href='%23l5c'/%3e%3c/animateMotion%3e%3canimate%20attributeName='opacity'%20values='0;0.9;0.9;0'%20keyTimes='0;0.1;0.85;1'%20dur='3s'%20begin='1.2s'%20repeatCount='indefinite'/%3e%3c/circle%3e%3c/svg%3e)
A hallucination is when an AI gives a confident answer that isn't supported by the source. In legal due diligence that's unacceptable, so the goal is to make the system hallucination-resistant by design: not a claim that it never errs, but built-in checks that catch unreliable output before it reaches you.
What is an AI hallucination?
A hallucination is when an AI produces a confident, plausible-sounding answer that isn't actually supported by the underlying source: a citation that doesn't exist, an obligation that was never in the contract, a clause subtly misread. The output looks right, which is exactly what makes it risky.
Why it matters in legal due diligence
In everyday drafting, a stray error is an inconvenience. In M&A due diligence, a missed or invented finding can shape a price, a warranty, or a decision to proceed. The tolerance for confident-but-wrong output is close to zero, which is why reliability, not raw capability, is the real test of legal AI.
What "hallucination-resistant by design" means
The honest goal is not a tool that never errs; no AI can promise that. It is a system engineered to catch unreliable output before it reaches you. In practice that means several techniques working together: multiple agents cross-examining each finding rather than a single pass; using different AI models for the tasks each handles best; confidence thresholds that separate the settled from the uncertain; outputs grounded in the source documents rather than the model's memory; and legal logic authored by experienced practitioners that constrains how the AI reasons. How well these hold up depends on the quality of the underlying AI engineering.
The role of traceability
Resistance is only half the answer; verifiability is the other. When every finding links back to its exact source paragraph, an error has nowhere to hide: a reviewer can check the claim against the document in seconds. Traceability turns trust into something you can test.
Why people stay in the loop
Because no system is perfect, the responsible design keeps a person in the loop: low-confidence findings are flagged for human review, and the deal team makes the final call. The combination, resistance by design, full traceability, and human judgement on the margin, is what makes AI findings usable on a real deal.
- ✓ A hallucination is confident output not supported by the source.
- ✓ "Hallucination-resistant by design" means built-in checks, not a claim of perfection.
- ✓ Cross-checking, confidence thresholds and human review lower the risk.
- ✓ Traceability lets you verify every finding against its source.