Skip to main content
Praxikon

Reporting an AI Act infringement as a whistleblower: Article 87 and the Dutch Whistleblower Protection Act

··26 min read

An employee sees at work that an AI application is being used that the AI Act prohibits, or that does not meet the rules. May they report it, to whom, and are they then protected against dismissal or other detriment? Article 87 of the AI Act answers in a single sentence: the EU Whistleblower Directive, Directive (EU) 2019/1937, applies to the reporting of infringements of the AI Act and to the protection of the people who report them. In the Netherlands the Directive was transposed by the Whistleblower Protection Act (Wet bescherming klokkenluiders, Wbk).

Suppose it is November 2026. A policy officer at a municipality notices that the customer contact centre uses software that infers from employees' voices and faces how stressed or irritated they are. What can she do? And what if the same thing happens at a bank? The examples are fictional.

In short

Article 87 reads: "Directive (EU) 2019/1937 shall apply to the reporting of infringements of this Regulation and the protection of persons reporting such infringements." The Digital Omnibus on AI (Regulation (EU) 2026/1744) did not amend it. It sits in Chapter IX and has applied since 2 August 2026 (Article 113).

The AI Act is not listed in the Annex to the Directive and does not amend that Annex. Article 87 brings reports about the AI Act under the Directive directly (recital 172). The draft Dutch AI Act Implementation Act states that the Directive, implemented in the Netherlands by the Wbk, applies to reports of infringements of the AI Act, and that Article 87 needs no implementation.

Protection covers anyone who reports in a work-related context: employees and civil servants, but also self-employed contractors, trainees, volunteers, job applicants and people working for a supplier. An employer that normally employs at least fifty workers must have an internal reporting procedure; for banks and insurers that threshold does not apply.

Which Dutch authority receives an external report about the AI Act is not yet settled: no market surveillance authority for the AI Act has been designated yet, and the Wbk does not mention the AI Act. The draft Dutch AI Act Implementation Act does already designate the AP, the AFM and DNB, which are already competent authorities under the Wbk; as long as the draft is not law, that is not settled. Where the infringement also concerns the GDPR, the Dutch Data Protection Authority (AP) is competent for that part. The Dutch Whistleblowers Authority (Huis voor Klokkenluiders) gives advice and can investigate.

This piece is about the whistleblower: the worker who reports an infringement from within their job. Complaints by citizens and customers go through Article 85; see market surveillance powers and the right to complain under Articles 74, 77, 78 and 85. The European AI Office's reporting channels are covered in the three reporting routes of AI Act enforcement. And organisations' own legal reporting duties, for a serious incident or a data breach, are covered in reporting serious incidents under Article 73 and notifying a data breach in an AI system.

What Article 87 does, and what it does not

The Directive protects people who report breaches falling within the scope of the Union acts in its Annex, in areas such as financial services, product safety and the protection of personal data (Article 2(1) of the Directive). The AI Act did not exist in 2019 and is therefore not in that Annex. Nor did the AI Act add itself to it: Articles 102 to 110 amend other acts, not the Directive. Article 87 settles it directly. Recital 172 gives the reason: persons acting as whistleblowers on infringements of the AI Act should be protected under Union law.

There are two things Article 87 does not do. It creates no duty to report: nobody is obliged to blow the whistle. And it does not say which authority receives a report; the Directive leaves that to the Member States (Article 11(1) of the Directive).

What counts as a breach follows from the Directive: acts or omissions that are unlawful or that defeat the object or purpose of the rules (Article 5, point 1, of the Directive). It covers reasonable suspicions about actual or potential breaches, including attempts to conceal them (Article 5, point 2, of the Directive). A reporting person does not need conclusive proof, but does need reasonable grounds to believe that the information was true and fell within the scope of the Directive (Article 6(1) of the Directive).

How the Dutch Whistleblower Protection Act handles this

The Wbk (wetten.overheid.nl, BWBR0037852) was called the Whistleblowers Authority Act until 2023 and has applied in its current form since 18 February 2023. It speaks of a "misstand" (wrongdoing) and has two kinds. The first is a breach, or risk of a breach, of Union law, insofar as it concerns Union acts within the material scope of Article 2 of the Directive. The second is a wrongdoing affecting the public interest: a breach of a statutory rule, or a danger to, for example, the safety of persons or the proper functioning of the public service, where the public interest is at stake (Article 1 Wbk).

The Wbk does not mention the AI Act anywhere. In our reading, an infringement of the AI Act has nonetheless fallen within the first kind since 2 August 2026, because Article 87 makes the Directive apply to the AI Act and the Wbk implements that Directive. The draft explanatory memorandum to the Dutch AI Act Implementation Act (section 4.5.5.1) proceeds on the same basis: the Directive, implemented in the Netherlands by the Wbk, applies to the reporting of infringements of the Regulation. It does not address the definition of a breach of Union law in Article 1 Wbk. For the reporting person that matters: for a breach of Union law there is no separate test of whether the public interest is at stake. If a court takes a different view, the second kind remains, and then the public interest must be at stake: not merely personal interests, but a pattern, a structural character, or something serious or extensive.

Who is protected?

The Directive protects reporting persons in the private and public sector who acquired information on breaches in a work-related context: workers and civil servants, self-employed persons, shareholders and board members, volunteers and trainees, and anyone working under the supervision of contractors and suppliers (Article 4(1) of the Directive). People who have left, or have yet to start, are also protected (Article 4(2) and (3) of the Directive). Protection extends to facilitators, to colleagues and relatives who may suffer retaliation, and to legal entities connected with the reporting person (Article 4(4) of the Directive; Article 17ec Wbk).

For an AI application this means that the developer at the supplier, the consultant configuring the system or the freelancer labelling the data can all be protected as reporting persons. A citizen or customer without a work relationship is not a whistleblower; they file a complaint.

Reporting internally: the reporting procedure

An employer that normally employs at least fifty workers adopts a procedure for reporting a suspected wrongdoing within the organisation (Article 2(1) Wbk). The procedure provides, among other things, that a worker can report in writing, by telephone or in a meeting, to which independent officer, that they may consult a confidential adviser, that they receive an acknowledgement within seven days and information on the assessment and follow-up within at most three months after that (Article 2(2) Wbk). The employer records every report and destroys the data once it is no longer needed (Article 2a Wbk). For private employers with 50 to 249 workers these new requirements have applied since 17 December 2023; before that, the old Article 2 applied to them, which had required a reporting procedure since 2016 (Article 21c Wbk).

The fifty-worker threshold does not apply to employers that fall within the Union acts in Parts I.B and II of the Annex to the Directive, such as banks and insurers (Article 2(3) Wbk; Article 8(4) of the Directive). Municipalities may share resources for receiving and investigating reports (Article 2(4) Wbk).

Reporting internally first costs nothing: the reporting person chooses whether to report internally or go straight to an external channel (Article 10 of the Directive). The Directive encourages internal reporting where the breach can be addressed effectively internally and the reporting person considers there is no risk of retaliation (Article 7(2) of the Directive).

Reporting externally: to which authority?

This is where the gap lies. The Wbk lists the authorities that receive and follow up reports, insofar as they are competent in the areas of Article 2 of the Directive: the ACM, the AFM, the Dutch Data Protection Authority, DNB, the Whistleblowers Authority, the Health and Youth Care Inspectorate, the Dutch Healthcare Authority and the nuclear safety authority ANVS, plus organisations designated by order in council or ministerial regulation (Article 2c Wbk). None of them has been designated as a market surveillance authority for the AI Act; the Netherlands does not have one yet. The draft AI Act Implementation Act does not amend the Wbk list; the transposition table in the draft memorandum says that Article 87 "by its nature" needs no implementation. Which Dutch authority receives and follows up a report about the AI Act itself is therefore not yet settled. The draft is not law; submission to the House of Representatives is planned for the second quarter of 2027 (Parliamentary Paper 26643, No 1554).

The draft does, however, already designate market surveillance authorities in Article 2.2: the AP, the AFM and DNB for the prohibited practices, the AP and the AFM for the transparency obligations of Chapter IV, including Article 50, the AP for, among others, Annex III points 1, 3, 4 and 5(a), and the AFM, DNB and the AP for Annex III point 5(b) and (c), with a further allocation across categories of operators by ministerial regulation (Article 2.2(2) of the draft). Because they are already listed in Article 2c Wbk, in our reading they would then also become the external channel for AI Act reports; as long as the draft is not law, that is not settled.

What is settled:

  • The Dutch Data Protection Authority accepts reports about breaches of the GDPR, the Police Data Act and the Judicial and Criminal Records Act, including the security of systems that process personal data. Many AI applications that breach the AI Act process personal data; for that part, the AP is the channel. The AP itself states that it can currently only investigate (very) serious reports, because it receives no funding for this task.
  • The AFM and DNB receive reports in the areas in which they are competent. For a financial institution, the market surveillance authority for high-risk systems directly connected with its financial services is in principle the financial supervisor, unless the Member State designates another authority (Article 74(6) and (7)); the draft names the AFM, DNB and the AP for this, with a further allocation by ministerial regulation (Article 2.2(2) of the draft).
  • The Whistleblowers Authority advises reporting persons on the steps to take and refers them to the right body (Article 3a(2) Wbk). Its investigations department can, on request, investigate a suspected wrongdoing and how the employer treated the reporting person (Article 3a(3) Wbk).
  • A wrong authority forwards the report. An authority that receives a report for which it is not competent forwards it, with the reporting person's prior consent (Article 2e(4) Wbk).
  • A later-designated authority is covered too. The rules for external reporting channels also apply to a public body that receives a report of a breach of Union law under a statutory task (Article 2j Wbk). In our reading that includes the future market surveillance authority for the AI Act, even if it is not named in Article 2c.

At EU level a reporting person can turn to the AI Office, but only regarding providers of general-purpose AI models (Article 88) and the AI systems for which the AI Office has exclusive competence (Article 75(1)). The Directive protects people who report to a Union institution or body under the same conditions as an external report (Article 6(4) of the Directive). When it comes to protection, the Wbk mentions only the employer, a competent authority and a body as referred to in Article 2j (Article 17e Wbk); how a Dutch court would bring a report to the AI Office within that is not yet settled. For a municipality or bank using an AI system, the AI Office is usually not the supervisor.

What protection the reporting person gets

A reporting person may not be subjected to detriment during and after the handling of the report, provided they had reasonable grounds to believe the reported information was true (Article 17e Wbk). Detriment includes dismissal or suspension, a negative assessment, transfer, intimidation and early termination of a contract, as well as threats or attempts (Article 17da Wbk). If detriment occurs, it is presumed to result from the report (Article 17eb Wbk). The reporting person is not liable for breaching a confidentiality restriction if they had reasonable grounds to believe the report was necessary to reveal the wrongdoing and reported in accordance with the conditions of the Act; medical confidentiality, legal professional privilege and national security take precedence (Article 17f Wbk). Any clause restricting the right to report is in principle void (Article 17h Wbk).

Public disclosure, for example to the press, is protected under stricter conditions: you first reported externally to a competent authority, whether or not after an internal report, and have reasonable grounds to believe the investigation is not making sufficient progress, or you have reasonable grounds to fear an imminent or manifest danger to the public interest, retaliation in the case of external reporting or a lack of effective remedy (Article 17ea Wbk; Article 15 of the Directive).

Where the GDPR comes in

A report about an AI application almost always contains personal data: of the reporting person, of the colleague or manager to whom the breach is attributed, and often of the people the AI system decides about. The Directive says that any processing in the context of reports is carried out in accordance with the GDPR, and that personal data which are manifestly not relevant are not collected or are deleted without undue delay (Article 17 of the Directive).

Confidentiality. Everyone involved in a report must keep confidential data secret. That includes the identity of the reporting person and of the person to whom the wrongdoing is attributed. The identity of the reporting person is not disclosed without their consent (Article 1a Wbk; Articles 16 and 22 of the Directive).

The person concerned. Anyone named in the report keeps their rights: an effective remedy and a fair trial, the presumption of innocence and the rights of defence, including the right to be heard (Article 22(1) of the Directive). Under the GDPR an organisation must in principle inform the data subject about data it did not obtain from them, but not insofar as doing so would seriously impair the objectives of the processing (Article 14(5)(b) GDPR). And a copy provided on an access request may not adversely affect the rights of others, such as the reporting person (Article 15(4) GDPR).

Legal basis and special categories. In our reading an employer that must have a reporting procedure under Article 2 Wbk processes the data in a report on the basis of a legal obligation (Article 6(1)(c) GDPR), insofar as that is necessary for the procedure and the register; for an employer without that obligation, or for processing that goes further, Article 6(1)(f) GDPR, or for a public-sector employer point (e), is the more obvious basis. The Wbk gives only the Whistleblowers Authority an exemption for special categories of data and criminal data (Article 1c Wbk); for an employer the ordinary rules of Article 9 and Article 10 GDPR and the Dutch GDPR Implementation Act apply.

Two reports, two routes. If the report concerns an AI application that leaks personal data, the organisation may also have its own duty to notify the breach. That is separate from the whistleblower report; see notifying a data breach in an AI system.

Main example: the municipality

Emotion recognition in the contact centre. The municipal policy officer from the introduction sees that the software infers emotions from employees' voices and faces. That is emotion recognition in the workplace, which is prohibited except for medical or safety reasons (Article 5(1)(f)). The prohibition has applied since 2 February 2025 (Article 113, third paragraph, point (a)). See emotion recognition in the workplace.

What can she do?

  1. Report internally. The municipality has more than fifty employees and therefore a reporting procedure. The report goes to the designated officer; she receives an acknowledgement within seven days and information on follow-up within three months after that.
  2. Report externally to the AP for the GDPR part. The system processes personal data of employees and, depending on how it is set up, also special categories of personal data, such as data concerning health or biometric data processed to uniquely identify a person, for which Article 9 GDPR lays down a prohibition with exceptions. For that part the AP is a competent authority under the Wbk. For the AI Act prohibition itself, no market surveillance authority has yet been designated in the Netherlands; the draft Implementation Act names the AP, the AFM and DNB for the prohibited practices.
  3. Seek advice. If she is unsure about the route or the protection, she can ask the Whistleblowers Authority for confidential advice.

A report about a prohibited practice made before 2 August 2026 did not, in our reading, fall under Article 87 yet, because that article did not apply at the time. Such a report could still be a wrongdoing affecting the public interest under the Wbk.

The risk score in social assistance. Another civil servant sees that the municipality uses a system that gives social assistance applications a risk score, and that caseworkers adopt the score without their own review. Such a system falls under Annex III, point 5(a). But the deployer's obligations, such as human oversight (Article 26), apply only from 2 December 2027 (Article 113, third paragraph, point (c)). If the system was already in use before that date, they apply only after a significant change, and a system intended to be used by public authorities must comply by 2 August 2030 at the latest (Article 111(2)). A report today is therefore usually not yet a report of an infringement of Article 26. It can concern the GDPR, for example a decision that is in fact automated (Article 22 GDPR) or a missing DPIA (Article 35 GDPR), and then the AP is competent. See also the obligations of deployers.

Example 2: the bank and the insurer

The chatbot without disclosure. An employee of a bank sees that the bank has put its own chatbot on its website that does not tell customers they are talking to an AI system, while that is not obvious to a customer. The bank developed the chatbot itself and is therefore the provider. The obligation to inform people that they are interacting with an AI system has applied since 2 August 2026 (Article 50(1)). This is therefore already an infringement covered by Article 87.

A bank always has an internal reporting procedure, whatever its headcount (Article 2(3) Wbk). Financial sector rules contain their own reporting rules for breaches of those sector rules; insofar as they apply, they take precedence (Article 3(1) of the Directive; Article 1b Wbk). In our reading they do not apply to an infringement of Article 50 of the AI Act. To which authority the employee reports the transparency failure externally is not yet settled: no market surveillance authority for Article 50 has been designated in the Netherlands; the draft Implementation Act names the AP and the AFM for Chapter IV.

The insurer's pricing model. An employee of an insurer sees that a pricing model for health insurance (Annex III, point 5(c)) is being put into use without a fundamental rights impact assessment. That assessment (Article 27) and the other high-risk obligations apply from 2 December 2027. Anyone reporting now is, in our reading, reporting a risk of a future infringement; the Directive also covers breaches that are "very likely to occur" (Article 5, point 2, of the Directive). For a financial institution, the market surveillance authority for high-risk systems directly connected with its financial services is in principle the financial supervisor, unless the Member State designates another authority (Article 74(6) and (7)).

Briefly on education and HR. A lecturer who sees a university of applied sciences using emotion recognition during online exams reports a prohibited practice (Article 5(1)(f)). A recruiter who sees a selection tool rejecting applicants without human review usually reports a GDPR matter first; the high-risk obligations for HR systems (Annex III, point 4) apply from 2 December 2027.

From when?

  • Article 87 sits in Chapter IX, Section 4. That chapter is not among the exceptions in Article 113, third paragraph, and has therefore applied since 2 August 2026. The Omnibus did not amend Article 87.
  • The prohibited practices of Article 5 have applied since 2 February 2025; the new prohibitions in Article 5(1), first subparagraph, points (ba) and (bb), and Article 5(1a) and (1b) apply from 2 December 2026 (Article 113, third paragraph, point (a)).
  • Article 50 has applied since 2 August 2026.
  • The high-risk obligations apply from 2 December 2027 for Annex III and from 2 August 2028 for Annex I (Article 113, third paragraph, point (c)), subject to the transitional rule in Article 111(2).
  • The Wbk has applied in its current form since 18 February 2023; for private employers with 50 to 249 workers the new reporting procedure requirements have applied since 17 December 2023; before that, the old Article 2 applied to them, which had required a reporting procedure since 2016 (Article 21c Wbk).
  • The Dutch AI Act Implementation Act is a draft; submission to the House of Representatives is planned for the second quarter of 2027. See the consultation on the Dutch AI Act Implementation Act.

What to do now

As an employer (municipality, bank, insurer):

  1. Review your reporting policy. Does its description of a wrongdoing include infringements of the AI Act, or only "laws and regulations" in general? Make explicit that a suspicion about an AI application can be reported.
  2. Connect the reporting policy to your AI register. The officer receiving reports must be able to see which system it is, who owns it and what role the organisation has.
  3. Settle the GDPR side. Legal basis, retention period, who has access to the register, and how you handle information and access requests from people named in a report. Involve the data protection officer.
  4. Do not promise more than the law. Confidentiality of identity is a duty; full anonymity is not, and in a small team a report may still be traceable.
  5. Be honest about external reporting. The Wbk requires you to provide information on external reporting to competent authorities (Article 2(5) Wbk). Name the AP for the GDPR, the sector supervisor where competent and the Whistleblowers Authority, and say that the authority for the AI Act itself has not yet been designated; the draft Implementation Act names the AP, the AFM and DNB.

As an employee in doubt: record what you saw yourself, when and in which system, report through the internal procedure if you do not fear retaliation, and ask the Whistleblowers Authority for confidential advice if you are unsure about the route.

Want to know which AI Act and GDPR obligations come into view for an application and what role your organisation has? With the case check you describe the application in a few questions.

Frequently asked questions about whistleblowers and the AI Act

Sources

EUR-Lex: Regulation (EU) 2024/1689 (AI Act), original text, recital 172 (12 July 2024, accessed 19 November 2026)
EUR-Lex: Regulation (EU) 2026/1744 (Digital Omnibus on AI); Article 87 not amended (8 July 2026, published 24 July 2026, accessed 19 November 2026)
EUR-Lex: Regulation (EU) 2016/679 (GDPR), Articles 6, 9, 10, 14, 15, 22 and 35 (27 April 2016, accessed 19 November 2026)
Huis voor Klokkenluiders: Ik vermoed een misstand (I suspect a wrongdoing) (accessed 19 November 2026)
European Commission: AI Act Whistleblower Tool (last updated 6 October 2026, accessed 19 November 2026)
House of Representatives: Digital Affairs planning letter, Parliamentary Paper 26643, No 1554 (31 August 2026, accessed 19 November 2026)

Newsletter

Every Tuesday, the AI Act week ahead in 5 minutes

A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.

Practical and short · No spam · One-click unsubscribe

Written by

Zahed Ashkara, freelance AI and privacy consultant and legal adviser

Zahed Ashkara

Freelance AI and privacy consultant and legal adviser · AI governance & safety

Zahed Ashkara is a freelance AI and privacy consultant and legal adviser for the GDPR and the EU AI Act. Organisations hire him as AI governance consultant, privacy counsel, external DPO or AI governance lead, have him carry out a DPIA, FRIA or AI Act classification, or book an in-company training. He focuses on AI governance and safety, including AI agents. He works exclusively as a freelancer, through Embed AI (trade name Praxikon). He is a Certified AI Compliance Officer (CAICO) and has been a member of the NEN standards committee Artificial Intelligence & Big Data since July 2026.

Follow on LinkedInView expert profile

Use these sources in your AI assistant too: Praxikon in Claude and other assistants

Real-World Stories

Discover how organizations apply this in practice

Related articles

Codes of conduct under Article 95 of the AI Act: voluntary rules for AI that is not high-risk

Article 95 of the AI Act lets organisations draw up codes of conduct on a voluntary basis: to apply the requirements for high-risk systems to AI that is not high-risk, and to set their own objectives on ethics, the environment, AI literacy, inclusive design and vulnerable groups. What such a code does and does not do, who may draw it up, how it differs from the codes of practice under Articles 50 and 56 and from codes of conduct under the GDPR. With a partnership of municipalities as the main example.

EU AI ActArticle 95Codes of conduct

Does the AI Act apply to AI in cars, aircraft, trains and ships? Articles 102 to 110 and Section B of Annex I

For AI that is a safety component of a car, a tractor, a drone, marine equipment, a train or, since the Omnibus, a machine, only a handful of articles of the AI Act itself apply. The requirements for high-risk AI systems come in through the sectoral legislation: the Commission has to take them into account when it adopts delegated and implementing acts there. What Article 2(2), Section B of Annex I and Articles 102 to 110 provide, what the Omnibus changed, what still applies, and what a municipality, a leasing company, a university of applied sciences and a transport operator need to do.

EU AI ActArticle 102Article 109