Meet Tsuno at Sirha Lyon 2027 · Stand S.57Learn more
EN FR
Try the demo
Menu

Team and management

AI at work: what changes for employers in 2027

From 2 December 2027, EU rules on high-risk AI cover hiring and staff management tools: employers will have to inform staff, oversee and keep logs.

Published on

In a restaurant office, a manager marks up a wall calendar while a colleague looks at a laptop.

For an employer, the EU AI Act (Regulation (EU) 2024/1689) arrives in three stages. Since 2025, some practices are banned, including reading employees' emotions, and employers must support their teams' AI literacy. Since 2 August 2026, transparency obligations apply. And from 2 December 2027, AI tools used to hire, allocate tasks or evaluate staff become high-risk systems: the employer using them will have to inform, oversee and keep records. This article gives general guidance under EU law, with French examples where national law matters, not legal advice: for your own situation, talk to your data protection officer or an employment lawyer.

Tuesday, 10 am, in front of the office calendar

Nadira runs three restaurants. In the office above the kitchen, she has taken the 2027 wall calendar down and pinned it to the flip chart. Thibault, her operations manager, is listing on his laptop everything the team uses: the job board that screens applications, the rota, the time clock, the assistant the managers query.

"One vendor told me everything changes in August 2026," Thibault says. "Another says it's been pushed back." Nadira circles 2 December 2027 in red marker. Next to it she writes: "Who does what, tool by tool." She wants to know what she has to do, and by when.

(Fictional scene and characters.)

The timeline, date by date

The regulation entered into force on 1 August 2024 and applies in stages. Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, adopted on 8 July 2026, published in the EU Official Journal on 24 July 2026 and in force since 27 July 2026, moved some of those stages.

DateWhat appliesWhat it means for a restaurant
2 February 2025Prohibited practices (Article 5) and AI literacy (Article 4)No reading of employees' emotions; explain the tools to the team
2 August 2025Penalty rules (Article 99)Fines for prohibited practices can be imposed
27 July 2026Digital Omnibus in forceHigh-risk rules postponed, Article 4 rewritten
2 August 2026Transparency obligations (Article 50)People must know when they are talking to an AI or exposed to emotion recognition
2 December 2027High-risk systems in Annex III, including employmentAI for hiring, task allocation and evaluation falls under Article 26
2 August 2028High-risk systems in Annex I (regulated products)Regulated products, such as machinery, that embed AI

What already applies

Prohibited practices

Since 2 February 2025, Article 5 has banned, among other things, using a system that infers a person's emotions in the workplace, except for medical or safety reasons. Fines can reach 35 million euros or 7% of worldwide turnover (whichever is lower for an SME). The details, with front-of-house and kitchen examples, are in AI and emotions at work: what has been banned since 2025.

AI literacy

Article 4 has also applied since 2 February 2025. The Omnibus rewrote it: providers and deployers take measures to support the AI literacy of their staff and of people using these systems on their behalf, without having to guarantee a specific level for each person. In a restaurant, that means explaining to the team what the tool does, what it does not do, and when to double-check its answer. Half an hour in a team meeting and a short sheet beat an e-learning module nobody opens.

Transparency

Article 50 has applied since 2 August 2026. It mainly binds providers: a system designed to interact with people must let them know they are dealing with an AI, unless that is obvious. Anyone using an emotion recognition system (on customers, since it is banned on employees) must inform the people exposed to it. The Commission published guidelines on these obligations on 20 July 2026.

What arrives on 2 December 2027

HR tools listed in Annex III

Annex III, point 4, classifies two families of AI systems as high-risk:

  • those intended for recruitment or selection: placing targeted job ads, analysing and filtering applications, evaluating candidates;
  • those intended to make decisions on working conditions, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate employees' performance and behaviour.

A rota built on declared availability is not the same as a tool that hands out tasks based on a behavioural profile. Article 6 also allows some listed systems to fall outside the high-risk category if they pose no significant risk: it is up to the provider to show this and document it.

Why 2027 and not 2026

These obligations were due to apply on 2 August 2026. The Omnibus postponed them to 2 December 2027. The delay covers high-risk systems only. It does not postpone the bans, nor transparency (apart from a grace period until 2 December 2026 for marking content produced by some systems already on the market), and it changes nothing in the GDPR or national employment law.

Provider or deployer

The regulation separates two roles. The provider develops the system and places it on the market: it is responsible for the product's conformity (documentation, assessment, marking). The deployer is whoever uses it under its own authority in a professional context. A restaurant screening applications with AI software is a deployer.

The deployer's obligations under Article 26

Article 26 sets out the list. In plain terms, for an employer:

  1. Use the tool in line with the provider's instructions for use.
  2. Assign human oversight to people with the necessary competence, training, authority and support. An approval given without reading does not meet that duty.
  3. Where it controls the input data, make sure it is relevant and sufficiently representative.
  4. Monitor how the system works, and alert the provider in case of risk or incident.
  5. Keep automatically generated logs for at least six months, where they are under its control.
  6. Before putting the system into service at the workplace, inform workers' representatives and the affected workers that they will be subject to it.
  7. Inform people, candidates included, that a high-risk system is used to make or help make a decision about them.
  8. Use the provider's information to carry out, where required, the data protection impact assessment under the GDPR.

National law adds its own layer. In France, for example, the Labour Code requires prior notice to each employee (L1222-4) and, from 50 staff, informing and consulting the works council before rollout (L2312-38). Both are covered in a microphone at work: what the law says in France. Other EU countries have their own rules.

The employer checklist for 2027

StepWhenWho
List every tool that contains AI: job ads, CV screening, rota, time clock, assistant, camerasNowManagement and operations
Check that none reads employees' or candidates' emotionsNowManagement, with the vendor
Ask each vendor in writing whether its tool falls under Annex III, point 4First half of 2027Purchasing or management
Name the people in charge of human oversight, and train themBefore December 2027Management and managers
Draft the information notice for employees and candidatesBefore going liveHR
Inform, and consult where required, employee representativesBefore the decisionManagement
Set up log retention and an incident processBefore December 2027Operations and vendor
Update the GDPR record of processing and, where needed, the impact assessmentBefore going liveDPO or privacy lead

No panic, no shrugging

  • Believing the delay wipes the slate clean. The 2025 bans and the 2026 transparency rules already apply.
  • Believing every tool with AI in it is high-risk. An assistant that answers a question about a procedure does not fall under point 4 just because it is used at work.
  • Treating compliance as the vendor's problem. The provider certifies its product; the employer answers for how it uses it.
  • Waiting until November 2027 to tell the team. In France, prior information is already required by law.
  • Mistaking an automatic sign-off for human oversight. Article 26 requires competent, trained people with the authority to overrule the tool.

What to measure

  • The share of your tools for which you hold a written answer from the vendor on Annex III.
  • The number of decisions about people (hiring, role, promotion) where a named person reviewed the tool's proposal.
  • The time between a new tool arriving and the team being informed.

Where Tsuno fits

Tsuno does not make you compliant. What the software does and does not do is published on the trust page. Tsuno connects dated facts (training completed, checklist done or not, incident, hours) and leaves the decision to the manager. It refuses to rate a person in place of their manager, to predict who will resign or to recommend who to dismiss.

In Tsuno, as from your own AI assistant, an action always follows the same path: the tool prepares it, summarises the effect, you confirm, then it executes and logs. Rights are checked again at the moment of confirmation, and the log keeps who, when and what, with the state before and after. San, the voice assistant for teams (San Voice, available as a demo), evaluates no one. Whether a tool is high-risk depends on how it is used: ask every vendor, us included. To see how it works on a real case, visit the Tsuno San page or book a demo.

Further reading

The full framework is in our pillar guide, AI, voice and privacy at work: the guide. The regulation's first red line is covered in AI and emotions at work: what has been banned since 2025. On human decision-making in HR, read interpreting without judging: the limit of AI in people management, and on employee representatives, should you consult employee representatives before using a voice assistant?. To prepare your questions for vendors, see 12 questions to ask before connecting an AI to your business.

Key takeaways

2 December 2027 adds obligations on top of those already in force, including the bans and transparency rules. Between now and then, list your tools, get their classification in writing and decide who, in your business, reviews before deciding.

Frequently asked questions

When do the EU AI Act's high-risk rules apply to HR tools?

From 2 December 2027 for the systems listed in Annex III, including point 4: recruitment, selection, decisions on the employment relationship, task allocation based on behaviour, and monitoring and evaluating performance. The original date was 2 August 2026.

What did the Digital Omnibus on AI change?

Regulation (EU) 2026/1744 of 8 July 2026, published in the EU Official Journal on 24 July 2026 and in force since 27 July 2026, postponed the obligations for Annex III high-risk systems to 2 December 2027 and rewrote Article 4 on AI literacy, among other changes. It did not remove the prohibited practices.

Is a restaurant using AI recruitment software covered by the AI Act?

Yes, as a deployer: a business that uses an AI system under its own authority in a professional context. The software vendor is the provider and is responsible for the product's conformity. The employer carries the usage obligations of Article 26.

What must an employer do before using a high-risk AI system at work?

Under Article 26 of the AI Act, it must inform workers' representatives and affected workers before putting the system into service, assign human oversight to competent people, follow the provider's instructions for use and keep the logs for at least six months. National employment law may add its own information and consultation rules.

Is AI training for employees mandatory under the AI Act?

Article 4 has applied since 2 February 2025. Since it was rewritten by Regulation (EU) 2026/1744, it requires providers and deployers to take measures to support the AI literacy of their staff, without requiring a specific level for each individual.