Courts keep sanctioning lawyers over citations nobody checked.
The model can be measured first.

Opinion 512 named the duties.

ABA Formal Opinion 512 put the verification duty in writing, and a firm running its own model for confidentiality reasons is left with a practical question. Verified against what? No firm can check everything, but it can find out where its model fabricates and check hardest there.

The instrument reads the model's internal signals while it answers the research patterns your attorneys actually use, on a model the firm hosts or designates. It maps where the model fabricates and how often, and every flagged answer arrives in full so the firm can verify it directly. The deliverable is a signed, dated screening measurement. Detection performance so far: 0.852 AUC on models the instrument had never seen, fabricated-entity discrimination, length-controlled, leave-one-model-out. Method paper DOI: 10.5281/zenodo.21365654.

No client files, no privileged material.

It never touches client files or privileged material. It needs representative question types, research patterns the way your attorneys actually phrase them, and a model endpoint the firm designates: a self-hosted model or a vLLM-class private-cloud deployment the firm controls. Private Azure OpenAI exposes probabilities on output tokens only, which isn't enough for the instrument to read. Consumer AI tools the firm doesn't control are out of scope.

What an engagement costs.

The LLM Validation Report is $9,500. Continuous Assurance is $6,500 per quarter for one model. The pre-deployment Model Selection Study is $4,500 for two candidates, up to $7,500 for four. Full terms are on the pricing page. No work begins without a signed engagement letter, and no fee is ever success-based.

Questions firms ask first