If a piece of software promises to tell from your employees' faces or voices whether they are tired, angry or unmotivated, do not buy it: the AI Act bans inferring people's emotions in the workplace, except for medical or safety reasons. For every other monitoring tool, from email to browsing logs, Italy already has Article 4 of the Workers' Statute and the GDPR, with or without artificial intelligence.
What the AI Act actually bans
Regulation (EU) 2024/1689 lists prohibited practices in Article 5. Point (f) of paragraph 1 concerns placing on the market, putting into service or using AI systems to infer the emotions of a person in the workplace and in education institutions, unless the system is intended for medical or safety reasons. The ban has applied since 2 February 2025.
The Commission's guidelines on prohibited practices, published on 4 February 2025, are not binding: the final interpretation belongs to the Court of Justice of the EU. But they show how the Commission reads the rule, and it pays to take them into account.
Penalties, under Article 99, go up to 35 million euros or 7% of worldwide annual turnover. For SMEs the lower of the two applies. For a small company the ceiling is therefore 7% of turnover: enough to hurt.
The medieval case: the bell they could not stand
Here the Professor allows himself a digression to Flanders and Artois, among the cloth towns of the thirteenth and fourteenth centuries. In those cities the working day was not decided by a clock but by a bell. The historian Peter Stabel reconstructed the system in a study published in 2014: in most Flemish and Artesian cloth cities, working hours were set by the city authorities and announced by bells.
In Ypres, for instance, an addition to the keure (the statute) of 1309 ordered journeyman weavers not to start before the "workers' bell" rang and to stop immediately after the evening one. Anyone who was late had their pay reduced in proportion.
And you will say: what excellent work organization! Someone thought otherwise. In 1367 in Thérouanne, a small bishop's seat near Saint-Omer, the bishop and the chapter abolished the use of the workers' bell. The reason, reported in Stabel's study, is that the ringing of that bell gave rise to scandal and division among workers, fullers and other craftsmen.
Well, your company today stands in that same town. Monitoring software is the new bell: it tells everyone when to start, when to stop, how long they sat idle. If people feel it as control they do not understand, the productivity gain risks vanishing into discord.
What is banned, what is allowed with rules, what is free
Not all monitoring is prohibited. You need a map:
| Tool | Where it stands |
|---|---|
| Software inferring emotions, stress or "engagement" from employees' faces, voices or expressions | Banned by the AI Act (Art. 5(1)(f)), except for medical or safety reasons |
| Cameras and systems allowing remote monitoring of workers | Allowed only for organizational and production needs, workplace safety and protection of company assets, with a union agreement or Labour Inspectorate authorization (Art. 4(1) Workers' Statute) |
| Tools the worker uses to do the job, and access and attendance recording | Outside the authorization (Art. 4(2)), but with a notice and GDPR compliance |
| Systematic logging of everyone's web browsing | The Data Protection Authority has fined it: see below |
The text of Article 4 of the Workers' Statute also says something that is often forgotten: the data collected may be used for all purposes related to the employment relationship, disciplinary ones included, only if the worker has been given adequate information on how the tools are used and how checks are carried out, and in compliance with privacy law.
The blanket monitoring the Garante fined
In 2021 the Italian Data Protection Authority (Garante) fined the Municipality of Bolzano 84,000 euros. The body logged the internet browsing of all its employees, about a thousand, with date, time, user, workstation and website visited, and kept the logs for 30 days. Disciplinary proceedings against an employee had arisen from that data.
The Garante found the systematic, preventive collection neither necessary nor proportionate for network security. It also noted notices scattered across several documents, the absence of an impact assessment, and that replacing names with the computer code was not enough to make the data anonymous. It ordered the existing logs to be deleted and limited future checks to genuine traffic anomalies.
It is not a private company, and AI had nothing to do with it. But the principle holds for any tool, even the most sophisticated: if monitoring is generalized, preventive and opaque, it is at risk.
How to think before buying a tool
- What does it really read? Ask the vendor whether the system processes faces, voice, keyboard or movements, and whether it produces assessments of mood, stress or attention. If so, stop.
- Why do you need it? Write the purpose in one sentence. "Knowing whether they are happy" is not an organizational need; "protecting a warehouse" maybe is.
- Who knows? If you need a union agreement or the Inspectorate, it takes weeks: do not install first.
- What is kept, and for how long? The less data you collect, the less risk. The Bolzano case grew out of indiscriminate retention.
- Who decides, based on the data? If an AI score ends up in an evaluation, a person must review it: we discuss this in If the AI gets it wrong, who pays?.
A 20-person company has no legal department: that is why caution counts double. If your labor consultant or firm follows you, involve them before purchase: AI for labor consultants explains how these professionals are facing the new rules. On the privacy side, GDPR and artificial intelligence is also useful.
What to do on Monday morning
- Inventory your monitoring. List every tool that records something about employees: cameras, email, browsing, management systems, company apps. Write down what it collects, for how long and who sees it.
- Ask vendors what the software does. One precise question: does it process face, voice or emotions? If the answer is yes or evasive, suspend its use until you have a written opinion.
- Update the notice. Explain in plain words what is monitored and why, and hand it to everyone. If you already use AI at work, pair it with internal rules like those in ChatGPT at work: rules for employees.
A bell in Thérouanne cost more discord than it gained in hours. The Professor prefers tools people understand and trust. If you want to put your company's monitoring in order without ending up in a dispute, book a free introductory call: in half an hour we look at what you use and what is worth changing. You will also find the services and the page for business owners.
The Professor of Penzai
