GuardX tests your AI and ML models requirement by requirement against the EU AI Act, SR 26-2, the NIST AI RMF and your own policies. It works from evidence your firm already holds, and it flags what's missing instead of guessing.
NEW The GuardX Frontier Index — free quarterly rankings of frontier models on financial regulationEach regulation is broken into testable requirements. Each requirement is checked against your model's evidence, and every result cites the clause it answers to.
Classify the model by use, jurisdiction and risk tier, and establish which regimes apply before any testing starts.
Turn each applicable regulation into specific, testable requirements, reconciled where regimes overlap so nothing is assessed twice.
Draw on what the firm already holds: model documentation, validation reports, data lineage, access and monitoring records. Missing records are flagged as gaps, never assumed.
A result for every requirement — met, gap or evidence missing — with the clause cited, an accountable owner and the remediation needed.
Outputs are built to be handed over, not reworked. Every conclusion traces back to a clause and a piece of evidence.
A matrix of every requirement by model, with status, cited clause and supporting evidence.
Each gap with an accountable owner, the remediation required and the regulation it closes against.
The records behind each finding, assembled for validation, internal audit or a supervisory request.
When a model or a rule changes, only the affected requirements are re-tested.
Every quarter, GuardX ranks leading frontier models against the regulations financial institutions operate under. Each model is scored on two things: whether it understands a regulation, and whether it adheres to it while carrying out real operational work. The rankings and methodology are free and open to everyone.
Can the model state what a rule requires, when it applies and where its limits are? Tested with questions grounded in regulation text and supervisory guidance.
Does the model follow the rule while doing the work? Tested on realistic tasks — KYC reviews, adverse action notices, complaint handling, credit memos — and scored on whether the output is compliant.
| Rank | Model | Understanding | Adherence | Overall |
|---|---|---|---|---|
| 1 | [Frontier model] | — | ||
| 2 | [Frontier model] | — | ||
| 3 | [Frontier model] | — |
An open benchmark for the whole industry: how frontier models compare on financial regulation, across every domain and both dimensions.
The index tells you how models compare in general. A deep dive tells you which model is fit for your use case, under your rules.
Large institutions run many models and must answer for all of them. GuardX keeps that answer in one place without halting the work.
Data, access, model and control records the firm already holds are read into a single risk picture for every model.
Findings go only to the functions the risk affects. Everyone else stays off the path, and the reason is recorded.
Each new release is compared with its last approval. Review reopens only for the requirements where the risk changed.
Bring a model or a use case and the regimes it falls under. We'll walk through what an assessment or deep dive would find.