Put a name against every AI system you use
Art50 keeps a record of who is responsible for AI in your business, system by system, versioned so you can show it on any date. Free forever.
No card. No time limit.
Who Is Responsible for the AI in Your Company?
It is the second question on most vendor forms, the first one an auditor asks, and the only one that matters the morning something goes wrong. Most companies cannot answer it with a name.
Ask a company which AI tools it uses and you will get a list. Ask who is responsible for AI inside that company and the room goes quiet, or somebody says “IT”, or somebody says “well, all of us really”.
Neither is an answer. Both are what people say when nobody has decided.
Why your clients ask who is responsible for AI
The question is not idle curiosity. It appears on AI vendor questionnaires, in security reviews and in procurement packs, because the company asking has obligations of its own and cannot discharge them by pointing at a supplier’s tooling. They need to point at a person.
Under the EU AI Act, obligations attach to organisations, and organisations act through people. When a regulator, a client or a journalist asks how a particular output came to be published, “the system did it” closes no conversations. Somebody approved the use. Somebody checked the result. The question is only ever which somebody.
The answer most companies actually give
In September we asked 66 companies how they keep track of where AI is used, and who maintains that record. The most common honest answer was that there is no record and no maintainer. One founder put it more clearly than any policy document would:
“The honest small-company answer is that the record is the founder’s working knowledge. I know every place AI touches our workflow because I approved each one.”
That is true, and at four people it is even reasonable. It stops working in three ordinary situations. The founder goes on holiday. A second person joins and starts using something nobody told the founder about. Or a client asks for the answer in writing, and working knowledge does not export.
A record that lives in one person’s head is not a record. It is a dependency.
What a good answer looks like
Three things separate an answer that satisfies a reviewer from one that produces a follow-up call.
It names a person, not a department. “IT owns it” fails because IT did not choose the tool, does not see its output, and cannot judge whether the output was any good. The marketing team picked it. Marketing owns it.
It gives a role as well as a name. A name proves a person exists. A role proves that person has the standing to stop the thing, change it or switch it off. Those are different claims, and the second is the one being tested.
It is per system, not per company. One “Head of AI” across eleven tools is a title, not an owner. The person who understands what your transcription tool does to client recordings is almost never the person who understands what your lead-scoring model does to inbound enquiries.
“Each system has a named owner. Content tools sit with our Head of Content, the scoring model sits with our CTO. The owner approves changes to how it is used and is the first point of contact if something goes wrong.”
Deciding who is responsible for AI, system by system
The useful test is not who understands the technology. It is who answers for the outcome.
If an AI-drafted email goes to a client with the wrong price in it, the person who has to make that right is the owner of the drafting tool. If a candidate is ranked badly by a scoring model, the owner is whoever has to explain the ranking, not whoever deployed the model. Follow the consequence and you find the owner, usually in one step.
For most small companies this produces a short and slightly uncomfortable list, because several lines come back to the same two people. That is fine. It is still better than a blank, and it makes the case for the next hire more concrete than any budget spreadsheet.
What changes once the name is written down
Three things, quickly.
The questionnaire becomes a copy-and-paste job instead of a week of asking around, which is itself a signal to the client about how the place is run.
The gaps become visible. A tool with nobody against it is not a paperwork problem, it is a tool nobody is watching, and you only find those by making a list and seeing which lines are empty.
And the question stops being frightening. Most of the fear around this comes from not knowing the answer, not from the answer being bad.
That list is what Art50 keeps: every AI system you use, what it is for, and the named person responsible for it. Records are versioned and append-only, so you can show who owned what on any date rather than only today. The free plan covers an inventory item and twenty records a month, with no card.
Art50 records human review of AI-assisted work and keeps an inventory of the AI a company uses. It provides evidence, not legal advice. Article 50 of the EU AI Act has applied since 2 August 2026; the full text of the article is here.
