Back to the blog

AI in hiring and HR gets more time: what the 7 May deal on the AI Act means

· By Bee Flow · 5 min read

On 7 May 2026 the European Parliament and the Council reached a provisional deal to move the high-risk rules of the AI Act, recruitment and staff management included. The dates as agreed, what an employer does in the meantime, and what stays the same.

A stack of application folders with a magnifying glass on top, next to a desk calendar where an amber arrow moves a pin to a later month

Early on Thursday 7 May 2026, negotiators for the European Parliament and the Council reached a provisional agreement on the Digital Omnibus on AI, the package that amends the AI Act (European Parliament press release). The part everyone is talking about is the delay. Under the law as it stands, the rules for so-called high-risk AI systems apply from 2 August 2026. Under the deal, that becomes 2 December 2027 for stand-alone systems and 2 August 2028 for AI built into products.

For HR the first date is the one that counts, because recruitment, selection and managing staff are on the high-risk list. Below: what the deal moves, what an employer sensibly does in the meantime, and what stays as it is. This is not legal advice.

Which HR software it covers

Annex III, point 4 of the AI Act names two groups (text of Annex III). The first is AI for recruitment and selection: placing targeted job ads, analysing and filtering applications, evaluating candidates. The second is AI used to make decisions on terms of employment, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate the performance and behaviour of staff.

Think of an applicant platform that ranks CVs, a scheduling tool that hands out shifts based on past performance, or software that scores productivity per employee. Features like these often sit in a module of a package you have used for years.

An employer that uses such a system is a deployer in the Act's terms. The vendor is the provider. The employer's duties are in Article 26 (text of Article 26): use the system according to its instructions for use, assign human oversight to people with the necessary competence, training and authority, monitor how it works, keep the automatically generated logs for at least six months, and, before putting it into use at work, inform workers' representatives and the affected staff that they will be subject to it.

What the deal moves

The Parliament's press release names employment explicitly among the stand-alone high-risk uses whose obligations would apply from 2 December 2027. As we read it, the Article 26 duties for such HR systems move with them, since they are part of the same high-risk obligations.

It is still a provisional agreement. Parliament and Council have to adopt it formally, and according to the same press release they intend to do so before 2 August 2026, the date the current high-risk rules start to apply. Until that adoption, the law still says 2 August 2026, and the final text has not been published. The Parliament's Legislative Train page tracks the next steps. Last November we wrote about the Commission's proposal this deal grew out of.

Our advice: plan for December 2027, but throw away nothing you prepared for August.

What to do in the meantime

Start with a list. Which software in recruitment, scheduling and appraisal uses AI, and for what exactly? Ask the teams themselves too, because an AI feature sometimes arrives switched on with an ordinary update.

Then put a few questions to the vendor in writing. Do they consider the product to fall under Annex III, point 4? If so, how do they plan to comply by the new date? Will you get instructions for use that explain how human oversight works? Which logs does the system create, and can you keep them for six months? An answer like "we will look at that once the text is final" tells you something as well.

Discuss the use with the works council or staff representatives. Dutch works council law often requires that already (see below), and the duty to inform in Article 26 will come on top. A conversation you have in 2026 is no surprise in 2027.

Tell staff and applicants in plain words what the system does and who looks at its output. And agree who that person is: someone allowed to overrule the software's suggestion, who knows when that is needed.

What does not change

Part of the AI Act already applies. Since 2 February 2025, AI that infers people's emotions in the workplace is banned, except for medical or safety reasons (Article 5(1)(f)). The deal leaves that ban in place and, according to the press release, adds a new one on AI that creates sexual images of identifiable people without their consent. We wrote about the emotion recognition ban in February.

The GDPR (AVG in Dutch) is not changed by this deal. It applies to the personal data of applicants and staff that such a system processes, including in the months until December 2027 (GDPR).

The Dutch Works Councils Act (Wet op de ondernemingsraden) stays as it is too. Its Article 27 requires the works council's consent for arrangements on processing and protecting staff personal data (point k) and for facilities aimed at or suitable for observing or checking staff attendance, behaviour or performance (point l) (statute text, in Dutch). Software that tracks performance or allocates tasks based on behaviour can fall under that, whatever date the AI Act ends up with. A works council is mandatory where at least fifty people normally work (Article 2).

Where Bee Flow helps

Next to those systems, an HR team does a lot of ordinary writing and looking things up: job ads, questions about leave rules, notes of a conversation. The same question comes up there: what happens to staff data? In Bee Flow, the Privacy Shield recognises names, addresses, Dutch citizen service numbers and other personal data before a message goes to a language model, and by default holds the message back. You choose the model, in the cloud, in a European data centre or on your own server, and a knowledge base holding your staff handbook answers with a link to the source. The privacy and security page explains how that works. If you would like to talk about whether it fits your HR work, get in touch.

#compliance #ai-at-work