An AI-assisted grievance can leave an employer with a case considerably wider than the employee’s original complaint. In this two-part series we consider what employers should do about it: first, how to establish what the employee is actually alleging; and, in Part 2, how to deal with misuse of the technology.
Part 1: “AI-generated grievances: who is really making the allegation?” — Friday 16 October 2026. Part 2: “Shadow AI and workplace grievances: when can employers take disciplinary action?” — Sunday 18 October 2026.
A grievance ought to tell an employer what happened in the employee’s own words. Now with AI tools being so readily accessible, it may arrive as a legal submission which the employee neither drafted nor fully understands. Whilst, the employer must respond to genuine concerns, it should not be expected to investigate every theory a machine can produce.
An employee may genuinely object to a performance meeting or a manager’s conduct. Feed that account into a generative AI tool, however, and it may return allegations of discrimination, victimisation or procedural impropriety which the employee never intended to make. Some may warrant investigation. Others may have no discernible factual basis.
The employer is left with both the grievance and the technology’s embellishments. That should not be the starting point for an investigation. The first task is to identify the actual complaint, not simply accept every legal characterisation in the document.
Irish employment law has yet to settle how internal procedures should deal with AI-written grievances. The response now emerging from the WRC and the courts is nevertheless instructive: the person advancing a factual account must be capable of standing over it. Employers should be prepared to apply that principle themselves.
What the WRC decisions actually show
In Oliveira v Ryanair DAC, ADJ-00055225 (1 October 2025), an unrepresented employee pursued equality complaints concerning race, family status, harassment and victimisation. At the first hearing, extensive documentation had been produced but the particulars needed to answer the complaints had not been identified. The adjudication officer adjourned the matter to allow the employee to clarify his case and the employer to respond. The subsequent submissions contained authorities which were irrelevant, misquoted or non-existent.
The adjudication officer criticised the use of ‘phantom citations’, describing the approach as ‘egregious and an abuse of process’. The complaints ultimately turned on the absence of the necessary proof, rather than on a general prohibition against AI. For employers, the concern is one of credibility: reliance on legal material (or any material for that matter) that does not exist invites scrutiny of the account in which it appears.
The problem of volume and particularity arose again in Goncalves v Valshan Unlimited, ADJ-00056185 (11 December 2025). The employer’s representatives challenged the reliability and apparent AI origins of extensive pre-hearing material, with the case file approaching 2,000 pages. The complainant herself acknowledged using AI to organise her case. The adjudicator experienced considerable difficulty identifying a coherent chronology and the factual basis of the individual complaints.
The adjudicator was concerned about the apparent absence of human oversight and struggled to isolate the chronology. She therefore sought the complainant’s own oral account. That is the important feature of the decision: a lengthy written case cannot substitute for evidence of what the complainant says actually happened. However, it does not follow that every AI-assisted submission was inadmissible.
In Ferko v Beyond Reach Limited trading as Car Wash Crew, ADJ-00060622 (2 March 2026) , the dispute was narrower. The employee alleged that his written employment terms had been furnished approximately a year late. His sister had prepared the WRC complaint form using AI, and he accepted that he was not fully familiar with its contents. The adjudication officer preferred the employer’s evidence about its induction procedure and the provision of written terms within days of commencement. The employee’s difficulty explaining and verifying the allegations reduced the weight of his evidence. The adjudicator observed that “a complainant remains responsible for the accuracy of allegations made on his behalf”.
In Ferko v Beyond Reach Limited trading as Car Wash Crew, ADJ-00060622 (2 March 2026), the dispute was narrower. The employee alleged that his written employment terms had been furnished approximately a year late. His sister had prepared the WRC complaint form using AI, and he accepted that he was not fully familiar with its contents. The adjudication officer preferred the employer’s evidence about its induction procedure and the provision of written terms within days of commencement. The employee’s difficulty explaining and verifying the allegations reduced the weight of his evidence. The adjudicator observed that “a complainant remains responsible for the accuracy of allegations made on his behalf”.
These cases do not make AI use a ground for rejecting a complaint. They illustrate a more practical problem: the quality of the drafting tells the employer very little about what the employee can actually establish. It is the facts, and the person prepared to stand over them, that matter.
The WRC’s approach: the submission remains the party’s responsibility
The WRC Guidance on the use of AI tools to prepare material for submission to the WRC was first issued on 20 October 2025 and updated on 26 May 2026. It is guidance for parties in WRC proceedings, not a statutory Code of Practice on the conduct of internal grievances. It states that the WRC treats submissions as the party’s own, that legal material must be checked, and that an individual may be asked to clarify what has been submitted. Disclosure of AI assistance is encouraged where appropriate, but it is not a universal mandatory declaration.
Investigate the event, not the legal labels
The practical danger is that HR begins investigating a document rather than an event. An AI tool can supply allegations, remedies and statutory labels which were never part of the employee’s original account. Before fixing the terms of reference, the employer should bring the discussion back to the underlying conduct.
Consider an employee who complains that their manager spoke aggressively during a performance meeting. After consulting an AI tool, the employee submits a formal grievance alleging bullying, discrimination, victimisation, psychological injury and breaches of several statutory obligations. The original concern may involve a single conversation.
Should every legal label trigger a separate investigation? We do not think so. The employer should ask what occurred, who may have witnessed it, and which alleged treatment warrants inquiry. The label alone cannot determine the investigation’s scope.
Under the Code of Practice on Grievance and Disciplinary Procedures, S.I. No. 146/2000, employee grievances must be “fairly examined and processed”. Fair examination does not mean accepting every assertion without clarification. Nor does it require the employer to adopt the legal characterisation chosen by an AI system. An employer should be prepared to say:
Please identify the incidents about which you complain, the conduct you say occurred and the witnesses or records which may assist. We will then identify the matters requiring investigation.
Require the employee to stand over the factual case
Grievance procedures should allow the employer, before commissioning a formal investigation, to request a short schedule of material allegations for the employee to confirm. This is a case-management measure for complaints which lack particularity, not an additional hurdle in every grievance.
The employee should identify what they witnessed, what they were told by others and what they have inferred. Where the document advances discrimination or other legal conclusions, the employer can ask which events are said to support them. The employee should also be free to correct any assertion generated by a drafting tool which they do not adopt. The employee need not possess legal expertise or forgo representation, interpretation or drafting assistance. The question is whether they understand and adopt the material allegations put forward in their name.
There is a real difference between saying, ‘My manager excluded me from a meeting because I was pregnant’, and submitting a sweeping allegation of systemic discrimination without being able to identify the conduct concerned. The first calls for inquiry. The second may first require particulars.
An employer need not commission a full inquiry into an assertion which remains unparticularised after reasonable requests for clarification. It should record that decision and the reasons for it, while continuing to address any identifiable discrimination, safety or protected-disclosure concern. Appropriate assistance and accommodations must be available during clarification.
The courts are insisting on a witness’s own evidence
Since 1 September 2026, High Court Practice Direction HC 142 (paragraphs 25–28) has required, unless the Court otherwise directs, a declaration in an opening paragraph of affidavits and witness statements intended to stand as evidence. Paragraph 27 requires the witness to confirm that generative AI was not used to generate the substantive evidence and that the document reflects the witness’s own knowledge and recollection. Paragraph 28 permits administrative assistance, including formatting and spell-checking.
Court of Appeal Practice Direction CA 16 adopts corresponding requirements. Circuit Court Practice Direction CC 34, in force since 21 September 2026, takes a different approach: paragraph 2 says substantive generative AI should not be used to prepare affidavits, statements of facts or evidence, apart from ancillary purposes. None of these directions governs an internal workplace grievance. Their common concern is that factual evidence remain attributable to the person advancing it.
There is no proposal here to turn an internal grievance into a sworn affidavit. The better course is to borrow the discipline behind the courts’ approach. Employers should be able, where appropriate, to ask employees to confirm the facts they advance, distinguish personal knowledge from second-hand information and identify substantive AI drafting assistance. A short declaration could read:
“I confirm that I have read this grievance and that it records the factual concerns I wish to raise. Facts stated as within my own knowledge are accurate to the best of my knowledge and belief. I have identified matters based on information received from others or about which I am uncertain. I [have / have not] used generative AI to assist with the substantive preparation of this grievance. Where AI was used, I have reviewed the document and confirm that I adopt the factual allegations made in my name.”
Refusal to sign should not automatically defeat a grievance. Nor should the declaration be treated as an empty formality. Where an employee has had a fair opportunity to explain or correct the account, an unexplained refusal to adopt material allegations may properly affect the weight attached to them. The investigator should record the reasons, consider independent evidence and allow for any relevant accommodation.
When the declaration proves inaccurate
A declaration serves a purpose only if it can be tested. Where an employee denies substantive AI assistance or adopts factual assertions later shown to be invented, the employer is entitled to put the discrepancy to them. A knowingly false declaration may affect the credibility of the account and raise a separate issue under the employer’s conduct rules.
Suspicion is a reason to enquire, not a finding of dishonesty. AI may have assisted with drafting even where the employee has a sound complaint. The question is what was materially inaccurate, what the employee knew and what remains to be investigated. Part 2 considers when separate disciplinary action may follow.
SPOTLIGHT | Build factual authentication into the procedure
The employer should preserve the grievance as submitted and then:
- set out the material factual allegations in a short schedule;
- distinguish first-hand knowledge from reports by others and inference; and
- obtain a signed or otherwise recorded confirmation of the account the employee adopts.
Retain that schedule, the employee’s corrections and the reasons for the investigation’s scope. If the account later changes, the record gives the employer a proper basis for testing credibility. It is considerably more useful than speculating about whether the prose ‘sounds like AI’.
Part 2, ‘Shadow AI and workplace grievances: when can employers take disciplinary action?’, is now available. It considers what employers may prohibit, and when unauthorised disclosure, a false declaration or deliberately obstructive conduct can justify action under disciplinary procedures.
Alastair Purdy LLP advises employers on grievances, investigations, protected disclosures and the fair enforcement of workplace AI and information-security rules.