EU register of AI systems (Article 71): beyond the inventory

Having an internal inventory of your artificial intelligence systems is necessary, but not sufficient. The EU AI Act goes a step further with its EU register of AI systems, set out in Article 71: a public European database where certain high-risk systems must be registered. In the regulated projects I lead, this obligation changes the way AI is documented from day one.

What the Article 71 register is

Article 71 of the EU AI Act creates a database managed by the European Commission where high-risk AI systems are registered before being placed on the market or put into service. The aim is transparency: so that authorities and citizens can know which high-risk systems are operating and who is responsible for them.

This is not a cosmetic formality. Registration requires the system, its purpose and its owner to be perfectly documented, which is only possible if a good internal governance base exists beforehand.

From the internal inventory to the EU register of AI systems

This is where the key difference lies. The AI asset inventory is your internal tool for knowing what you have. The Article 71 register is the public, mandatory face for high-risk systems. One feeds the other: without a solid inventory, preparing the EU register of AI systems is a nightmare.

That is why I recommend designing the internal inventory with the fields the European register will ask for already in mind. That way, when it is time to register, the information is already structured and validated.

Who must register and what information is required

Those responsible for registration

The obligation falls mainly on providers of high-risk AI systems and, in certain cases, on users that are public entities. Identifying who is responsible for each system is the first control I cross-check with the EU AI Act checklist.

Data to register

The register requires information about the system’s purpose, its provider, the status of the conformity assessment and contact details, among others. Keeping that information up to date requires treating the EU register of AI systems as a living process, not a one-off registration.

How I prepare it in regulated projects

I integrate the register into governance from the design of the system. That logic of registering, classifying and keeping every AI system up to date is exactly what underpins the compliance tool I am developing, because doing it by hand for dozens of systems is not viable. The key is that the register should not be a last-minute race, but the natural reflection of good governance.

Mistakes I see when preparing the register

The most common mistake is confusing the Article 71 register with a simple list of applications. Noting down names is not enough: the European database asks for the system’s purpose, risk level, the owner’s contact details and an understandable description of how it works. When an organisation comes late to this, it discovers that gathering that information against the clock is far more expensive than having kept it up to date from the start.

Another frequent failing is treating the register as a fixed snapshot. AI systems change: they are retrained, change provider or expand their use. If the register is not updated when that happens, it stops reflecting reality and loses all its value as a control tool. That is why I always connect it with the obligations of general-purpose models, since many registered systems integrate precisely this type of model and inherit their transparency requirements.

Register and governance: two sides of the same coin

In my experience, the companies that handle the register best are those that already have a minimum of AI governance. It is no coincidence. The register forces you to answer questions that every mature organisation should have resolved: who owns each system, what decisions it automates and what happens if it fails. If those answers exist, registering the system is almost a formality; if they do not, the register exposes a governance gap that should be closed as soon as possible.

My recommendation is not to wait for the deadline. Start with the high-risk systems, document the essentials and create a routine to keep the register alive. Turning this obligation into an orderly process, rather than a last-minute race, transforms a regulatory burden into a sign of seriousness to clients and regulators.

In practice, the register of high-risk systems works as a mirror of the organisation’s internal maturity. To register a system rigorously you already need to have documented its purpose, its training data, its oversight measures and its assessment results. That is why I advise against waiting until the last moment: if you prepare that documentation while you develop, the register is an orderly formality; if you leave it to the end, it becomes a race against the clock that brings to light everything that was not well governed.

Conclusion: transparency as both obligation and advantage

The EU register of AI systems turns transparency into a legal obligation. Far from being a burden, it is an opportunity to demonstrate maturity: those who have their AI well documented and registered convey a trust that the market and regulators value more and more.

Frequently asked questions about the EU register of AI systems