The AI Act: Recording Work Is Not the Same as Judging It

A site manager reads a headline about the EU AI Act on his phone during a tea break, and a small worry lands somewhere behind his sternum. Does that regulation touch the app his cleaning crews tap in and out of every morning? Is a clock-in tool now, by some quirk of Brussels drafting, a “high-risk AI system”? The question sounds overcautious until you actually open the text, because the AI Act does single out software that manages people at work, and it puts a slice of it in the same risk tier as AI used in medical devices and power grids. The worry is reasonable. The answer, though, depends entirely on what the software actually does, not on whether it happens to run on a phone with a green icon.

What the AI Act actually flags

Regulation (EU) 2024/1689, the AI Act, lists categories of “high-risk” AI in its Annex III, and point 4 of that annex is dedicated to employment, workers’ management and access to self-employment. It covers AI built to recruit or select people, including systems that place targeted job adverts, filter applications or score candidates. It then goes further, into systems used to decide on promotion, termination or the allocation of tasks based on someone’s individual behaviour or personal traits, and into systems that monitor and evaluate the performance and conduct of people already in work. None of that is vague. It is a list of AI that makes or heavily shapes decisions about a human being’s job, and Brussels decided those decisions deserve scrutiny, human oversight and a paper trail, not a black box.

This is the branch of technology sometimes called algorithmic management: software that assigns shifts, ranks output, flags underperformers or nudges someone towards the door, with a manager rubber-stamping whatever the model concludes. It has crept into logistics, call centres and gig platforms over the last decade, usually sold as efficiency and arriving as something closer to a supervisor nobody elected. The AI Act does not ban it outright. What it does is force it into the open: providers and employers who deploy this kind of system face a fundamental rights impact assessment, a duty to keep logs, a requirement for competent human oversight, and an obligation to tell workers and their representatives, before the system goes live, that it is being used on them at all.

Where the line actually sits

Here is the distinction that gets lost every time this subject comes up in a trade press headline. There is a wide gap between an AI system that decides something about a worker and a record that simply states a fact about them. A tool that scores candidates, ranks performance or flags who should be let go is making a judgement, and the AI Act is right to treat judgement of that kind as high stakes. A tool that logs who clocked in at a job site, at what time, and for how many hours, is not judging anything. It is not weighing personal traits, it is not predicting who will underperform next quarter, and it is not deciding who gets the next shift. It states a fact that already happened, and it leaves every decision about what to do with that fact in human hands. Annex III catches the first kind of system. It has no reason to catch the second.

The practical test, for an owner running crews of cleaners, security guards or installers, is not whether software touches employment at all. It is whether that software makes an automated decision or produces an automated score about a person, and whether the worker can see, understand and challenge what it records about them. Transparency is the hinge the whole regulation turns on. An employer deploying a genuinely high-risk system under the AI Act has to inform workers’ representatives and the people affected before switching it on, has to keep records a human can audit, and has to make sure a competent person, not the algorithm, carries the final call. A tool that never makes that call in the first place has already cleared the hardest part of that obligation, simply by design. Recording is not deciding, and the regulation was never really written to catch the record.

Where this leaves a UK business

Worth being straightforward about one thing here: the United Kingdom is not bound by the EU AI Act. It left the sphere that regulation covers, and it has instead pursued its own pro-innovation, sector-led approach, with existing regulators such as the Information Commissioner’s Office folding AI oversight into the rules they already enforce rather than a single new statute. That does not mean UK firms can shrug the subject off. The ICO has been paying close attention to automated decision-making in recruitment and performance management under UK data protection law, and any UK company with staff, contracts or clients inside the EU can find the AI Act reaching across the Channel anyway, the moment its systems process people or produce effects there. The regulation aimed at Brussels rarely stays neatly inside its own borders once a business trades across them.

So the sensible move for a UK owner is not to wait and see which statute eventually lands on his desk. It is to look, right now, at what his own tools actually do to the people on his payroll. A rota app that quietly scores workers against each other and feeds that score into who gets more hours is drifting towards exactly the territory both the EU and the ICO are circling. A record that simply states where someone was and for how long is not, and staying on that side of the line is the cheaper, calmer place to run a business from regardless of which regulator eventually knocks.

In principle, then, the safest kind of workplace tool right now is the boring one: a transparent, geo-timestamped record that the worker himself can open and read, with no score attached and no automated verdict hiding behind it. GeoTapp was built along exactly that line. It does not profile anyone and it does not grade anyone; it is a geo-timestamped clock-in that the worker sees on his own phone, a statement of fact rather than a judgement, which is precisely why it sits outside the high-risk territory the AI Act was written for.

So before the next headline about the AI Act sends a jolt through your coffee break, it is worth asking a plainer question about your own workplace tools: does yours decide something about your people, or does it simply tell the truth about their day?