Spectralgraph
Law firms

The sanctions were for citations nobody verified. Measure first.

Courts keep sanctioning lawyers for AI-fabricated citations, and ABA Formal Opinion 512 has made verification a named duty. Spectralgraph is an independent AI-testing laboratory that measures where the model your firm hosts fabricates, so verification effort goes where the risk is.

Request the sample report See the public index
FIXED PRICE · FIVE BUSINESS DAYS · NO MEETINGS, EVERYTHING IN WRITING · NO CUSTOMER DATA
The professional moment

Opinion 512 named the duties. The sanctions lists keep growing.

The pattern in the sanctions orders is consistent: confident output, plausible citations, nobody checked. ABA Formal Opinion 512 frames the duties, competence, confidentiality, and verification, and firms that host their own models for confidentiality reasons now face a practical question: verified against what? A firm cannot verify everything; it can know where its model fabricates and verify there hardest.

It's not a checklist or a vendor questionnaire. What it does is measure the model itself, reading the model's own internal signals rather than grading its response output. Using your real question types on a model you host, it maps where the model fabricates, how often, and on which topics, down to specific flagged answers your team can verify with their own eyes. The deliverable is a signed, dated screening measurement built for your governance file. In testing so far it catches fabricated answers at about 0.85 AUC on models it has never seen before; that figure is provisional until the work is published.

What it needs from you

No client files. No privileged material. Five business days.

The test needs representative question types, research patterns as your attorneys actually phrase them, and a model endpoint the firm designates. It never touches client files or privileged material. In scope are models the firm hosts or controls, including private Azure OpenAI deployments; consumer AI tools the firm does not control are out of scope, and we say so.

The engagement is written-first and async by design, which tends to suit how firms prefer their vendors: everything answered in writing, on the record.

Terms

Fixed fees, written engagement, nothing success-based.

Fees are fixed and never success-based: the LLM Validation Report is $9,500, Continuous Assurance is $6,500 per quarter for one model, and the pre-deployment Model Selection Study runs $4,500–$7,500 by candidate count. Full details and the founding client program are on the home page. Every engagement runs under a written agreement before any work begins.

Straight answers

Questions firms ask first

Does the test see client files or privileged material?
Never. The instrument needs representative question types, the shapes of questions your attorneys actually ask, and a model endpoint the firm designates. Client files and privileged material stay inside the firm.
How does this relate to ABA Formal Opinion 512?
Opinion 512 frames a lawyer's duties when using generative AI: competence, confidentiality, and verification of output among them. The measurement does not discharge those duties, but it tells the firm where its model fabricates and on which topics, so verification effort lands where the risk actually is instead of everywhere at once.
Why do confidentiality-minded firms self-host, and does that change the test?
Firms increasingly run models they control precisely so client matter never leaves the building. That is exactly the deployment the instrument was built for: it measures a model you host or control, including private Azure OpenAI deployments, at an endpoint you designate.
Can the results help after a citation incident?
A measurement after an incident establishes the current fabrication baseline and, re-measured quarterly, documents improvement. What it cannot do is undo a filing; the cheaper moment to measure is before the sanction, not after.

See exactly what the firm would receive.

Reply and we will send the sample validation report, the same signed, dated document your risk committee would review.

Email the lab
REPLIES SAME DAY · EVERYTHING ANSWERED IN WRITING, ON THE RECORD