Spectralgraph
ISO/IEC 42001

Your AIMS needs evidence about the model. Not just about the process.

ISO/IEC 42001 certification covers the management system you run around your AI. It does not measure the model. Spectralgraph is an independent AI-testing laboratory, and measuring the model is the whole of what we do.

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

A certification confirms the process. The model itself is a separate job.

Accredited bodies are certifying AI management systems now, and the evidence behind those files has to come from somewhere. A certification confirms your governance operates. It cannot measure your model, and that is the piece we work on, from outside your organization.

We run your real question types against a model you host or control and read the model's internal signals while it answers. The report shows where it fabricates, how often, on which topics, and names the flagged answers so your own people can check them. What you get is a signed, dated screening measurement. Published performance: 0.852 AUC on models the instrument had never seen (fabricated-entity discrimination, length-controlled, leave-one-model-out), in the lab's method paper (DOI: 10.5281/zenodo.21365654).

What it needs from you

Two things: your question types and an endpoint you designate.

Nothing gets installed, and the test never touches customer or user data. What we can measure is a model you host or control whose deployment exposes prompt-side token probabilities, such as self-hosted open-weight models, vLLM-class private-cloud deployments, and fine-tunes. Private Azure OpenAI exposes probabilities on output tokens only, so it takes behavioral testing rather than the instrument-grade measurement. The first measurement is a dated baseline, and quarterly re-measurement builds the trend record behind it.

The same artifact serves NIST AI RMF programs as MEASURE-function evidence, produced independently. The MEASURE crosswalk goes subcategory by subcategory, including the seven it does not touch.

Terms

Fixed fees, never success-based.

The LLM Validation Report is $9,500. Continuous Assurance is $6,500 per quarter for one model. The pre-deployment Model Selection Study runs $4,500–$7,500 by candidate count. The rest is on the pricing page, and no work begins without a signed engagement letter.

Straight answers

Questions AIMS owners ask first

Does this certify us against ISO/IEC 42001?
No. Spectralgraph is not a certification body, is not accredited to ISO/IEC 42001, and does not perform certifications of any kind. What we hand you is model-level evidence: an independent, dated measurement of how your model behaves.
Where does the measurement fit in an AIMS?
That call is yours. AIMS owners generally file it under performance evaluation and monitoring: a dated baseline at adoption, then a quarterly trend record as models and prompts change. It lines up with the MEASURE function of the NIST AI RMF as well.
What does the test need access to?
Representative question types and a model endpoint you designate. It never touches customer or user data, and nothing is installed in your environment. What we can measure is a model you host or control whose deployment exposes prompt-side token probabilities, such as self-hosted open-weight models, vLLM-class private-cloud deployments, and fine-tunes. Private Azure OpenAI exposes probabilities on output tokens only, so it takes behavioral testing rather than the instrument-grade measurement.
Our certification body already reviews our AI governance. Why add measurement?
They answer different questions. The review confirms your management system exists and operates. The measurement shows how the model itself behaves, and a management-system certification isn't going to produce that number for you.

Look at the report before you decide anything.

Email the lab and we will send the sample validation report, so you can see the exact artifact your evidence file would hold.

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