The risk tier is derived, never chosen
A register where somebody picks "low" from a dropdown tells you what that person felt that morning. Ours computes the tier from what the system actually does — and it moves by itself when the answers change.
Platform capability
Nobody can name all the AI they run.
Ask any organisation to list the AI it runs and you will get the two or three systems somebody bought on purpose. The rest arrived inside products you already pay for, and nobody decided anything about it.
DerivedRisk tier, never typed
EU AI ActRole and class recorded
ISO 42001Controls behind it
OwnedEvery entry has a name
The boring half nobody has
Every compliance vendor now sells AI governance, and for most of them it is a policy template and a questionnaire. Both are useful. Neither answers the question an incident will ask first, which is simply: what AI do we run, what does it touch, and who said that was acceptable?
The awkward part is that the majority of an organisation's AI exposure is not the model somebody procured. It is the summarisation feature in the helpdesk, the drafting assistant in the CRM, the code completion in the IDE, the meeting transcription nobody switched off. Each one reads real data, most were enabled by default, and almost none went through a review — because nobody thinks of a product they already bought as an AI system.
So this is a register, and it is deliberately the unglamorous half. It records what is in use, what data reaches it, how much it does without a person, who owns it, and where it sits under the EU AI Act. Then it computes the risk rather than asking somebody to feel it.
Coverage
What it records, and what it works out
The inputs are facts about the system. The risk tier is a consequence.
What data reaches it
Personal, sensitive, health, financial, credentials, source code, customer content. The single largest factor, and the one people have not thought about when a tool was enabled by default.
How much it does unattended
Assistive, automated or agentic. An agent that chooses its own steps and calls tools can do damage an assistant cannot, and the tier reflects that.
Whether output lands on a person
Hiring, credit, pricing, access, moderation, a medical or legal suggestion. This is the factor regulators care about most, and it is a question most registers never ask.
Whether anyone can intervene
Human oversight lowers the tier — but never far enough to make an agentic system handling sensitive data look safe, because oversight after the fact is not the same as a person deciding.
Whether the vendor trains on your input
An unbounded disclosure of whatever you send. Weighted against what that actually is: public marketing copy is a nuisance, customer content is a contractual problem.
Your EU AI Act position
Role — provider, deployer or both — and risk class, with the reasoning recorded. Most organisations are deployers of far more AI than they provide, and the deployer obligations are the ones almost nobody has read.
Controls behind it, not beside it
The register connects to the AI assurance clauses you actually track — ISO 42001, the NIST AI RMF and the EU AI Act are all in the control library — so it is a programme rather than a list.
Approach
How to fill it without a six-month project
Start with what you already pay for.
-
01 · List the products, not the models
Go through the SaaS you already buy and record the AI features in each. This is the pass that finds the entries nobody expected, and it takes an afternoon.
-
02 · Answer the factual questions
What data reaches it, how autonomously it acts, whether output affects anybody, whether a person can intervene. The tier computes itself as you go.
-
03 · Classify the ones that matter
Work through the EU AI Act position for anything with an EU nexus, and record the reasoning. A documented determination that a system is not high-risk is itself something you have to keep.
-
04 · Approve, own and review
Every entry gets a named owner and an approval. Approved systems come back for review annually, because what they are connected to changes faster than the approval does.
Deliverables
What you actually receive.
The report is the product. If it cannot be acted on by a developer and understood by a director, we have not finished.
A register you can hand to an auditor
Every AI system with its owner, its data, its classification and its approval on record.
A derived risk position
Computed from what each system does, and re-computed whenever that changes.
The combinations that need attention
Agentic with nobody watching, personal data to a third party with no DPA, in use with no approval, overdue its review — flagged without being asked.
AI assurance controls, connected
The ISO 42001, NIST AI RMF and EU AI Act clauses you track, with their current state.
Is this for you?
Talk to us if any of these are true.
If none of them are, say so on the call and we will tell you honestly whether this is the right piece of work — or point you at the one that is.
Book a scoping call- You cannot produce a list of the AI your organisation uses.
- An agent in one of your products can call tools, write data or spend money.
- A customer, insurer or regulator has asked how you govern AI.
- Somebody enabled an AI feature in a tool that holds customer data, and no one reviewed it.
- You are preparing for ISO 42001 and have no inventory to build the ISMS scope on.
- You need to establish your EU AI Act position and do not know where to start.
How we work
Six steps, and no surprises.
The same engagement model applies to every piece of work we take on, so you always know what happens next.
01
Scope
A 30-minute call, then a written scope: what is in, what is out, what we need from you and what it costs. Nothing starts before you sign it.
02
Authorise
Rules of engagement, testing windows, escalation contacts and a signed authorisation. Out-of-hours windows where production cannot take the load.
03
Test
Automated coverage first, then manual testing where judgement is required. Critical findings are reported the day we confirm them, not at the end.
04
Report
One report a developer can act on and an executive can read, with evidence, reproduction steps, business impact and a fix for every finding.
05
Remediate
A walkthrough call with your engineers. We answer questions on the fix, not just the finding.
06
Retest
A free retest cycle to confirm the fixes hold, and a clean summary you can hand to a customer, auditor or board.
Questions
AI Register — answered.
The questions clients actually ask during scoping. If yours is not here, ask it directly.
Does it discover AI automatically?
No, and we are explicit about that rather than implying otherwise. This is a register: it records what has been declared. Automatic discovery — from OAuth grants and endpoint inventory — is a separate piece of work, and shipping a half-built discovery alongside the register would make the register's numbers untrustworthy from day one. What the register does do is make the declaring cheap enough that it actually happens.
Why can we not set the risk tier ourselves?
Because a tier somebody typed is a tier somebody typed. It records a judgement made on one day, by one person, and it does not change when the system does. A tier computed from the data, the autonomy, the human impact and the oversight tells you something — and when a team switches a tool from assistive to agentic, the register notices without anybody remembering to update it.
How does this differ from your AI security testing?
Different question entirely. AI red teaming asks whether a specific system can be made to misbehave — prompt injection, agent abuse, retrieval leakage — and it is an engagement. The register asks what you run, what it touches and who owns it, and it is a module. Most organisations need the register first, because you cannot test what you have not listed.
Is this legal advice on the EU AI Act?
No. The register records your position and the reasoning behind it so the determination exists and can be produced. It will tell you when a combination looks worth a proper look — high-risk with no human oversight recorded, say — but the classification is yours to make, with advice from someone qualified to give it.
What if a system turns out to be a prohibited practice?
The register will not let you approve it into use. That is the one place this module refuses rather than warns: there is no compliance route for a prohibited practice, so it is not something to control, it is something to stop. If the classification is wrong, correct it first.
Next step
Get a written scope and a fixed price.
A 30-minute call, then a scope document with what is in, what is out and what it costs. No obligation, and no charge for the conversation.