The recent Workplace Relations Commission (“WRC”) decision in Oliveira v Ryanair serves as a cautionary tale in the context of employment law claims – while AI can draft legal submissions, it cannot discharge the legal burden of proof or be a substitute for genuine engagement with evidence.
While the sums at stake were relatively modest at around €2,300 claimed by a former cabin crew member, with the WRC ultimately finding that he had been overpaid and still owed Ryanair €700, this decision’s significance lies in how the adjudication officer assessed the use of AI in preparing the employee’s submission.
So what happened?
Mr Oliveira, the employee, brought a claim to the WRC alleging unpaid wages, unlawful deductions and controversy around sick pay and annual leave following his resignation in June 2025. Ryanair responded with payslips and three separate explanations issued over a seven-month period setting out why certain payments were made in error and why the annual leave clawback was contractually provided for.
The adjudication officer found that Mr Oliveira had “erroneously” relied on AI in preparing his submissions and, critically, “did not engage with the substance of what was put to him” by Ryanair’s legal representatives. The adjudication officer noted that the employee failed to explain the key aspects of his claim, e.g. why he believed he was entitled to a further month’s salary in July when he had resigned and his employment had ended on 11 June 2025, or how he addressed Ryanair’s explanation of the annual leave rate. When challenged on these points during the hearing the employee could not offer an explanation such that the Adjudication Officer “formed the view that his submission had been prepared by AI”.
The Adjudication Officer concluded:
“it is both extraordinary and rare, in my experience, for a respondent to engage with a complainant’s objections in the degree of sustained, itemised detail, that they did in this case and the complainant did not engage with it in turn, at hearing or at any stage prior to it, preferring in my view, to rely instead on what AI erroneously told him.”
The claim was dismissed and the WRC found Mr Oliveira had in fact been overpaid.
How Ryanair’s defence shaped the Adjudicator’s view on AI
The Adjudication Officer’s pointed observation that the employee appeared to have relied on what AI “erroneously told him” is one of the more direct statements to date in Irish employment law about the dangers of deploying AI-generated legal submissions. Ryanair’s approach, which was grounded in a paper trail of evidence including payslips, contractual terms and a clear chronology of communications (which the Adjudicator described as “arithmetically reconciled and exhaustively detailed explanations” ) created a framework for which the complainant’s AI-generated submissions could not engage meaningfully. The observation that the complainant “did not offer an explanation in his own terms” alludes to the generic and disconnected nature of the submission in relation to the contractual issues raised.
What does this mean for employers?
For employers, this decision underscores the importance of having a well-structured and documented defence that will not only rebut a claim but also serve to expose the limitations of AI-reliant opponents against a backdrop of a significant uptick in the use of AI to drive internal grievances and legal claims. Given the rapid increase in the use of AI, adjudicators are increasingly emphasising credibility and the ability of parties to address the other side’s evidence directly – effectively calling out the use of AI as a smokescreen in addressing legal complaints properly. In instances such as this case where the disparity between written submissions and oral evidence is vast, the imbalance can be decisive.
A wider lens – recent Irish and UK caselaw
This decision fits neatly into a broader trend. In the January 2026 WRC case of Maclou v Kubicle, a HR professional bringing a protected disclosures claim, admitted to using AI in preparing her legal submissions. The employer’s legal representatives pointed out that some cases cited were misquoted, and others did not exist. The adjudication officer directed the complainant to file a concise, corrected submission recognising the unfairness on the other side in sending them “on a wild-goose chase for cases that don’t exist”. In an earlier separate Ryanair case, with the same complainant as above, the adjudicator dismissed the complainant’s AI generated submission calling it an “abuse of process”. Please see our previous article where we have discussed this case in more detail.
Other WRC decisions echo this tone, ultimately leading to the Court of Appeal weighing in twice this year and judges emphasising that litigants remain responsible for verifying AI generated content, regardless of whether errors stem from AI hallucination. One ruling set out clear principles: parties may use AI for research but must verify every citation, disclose AI use and ultimately take ownership of the final submission to the court.
The WRC and the Labour Court have updated their guidance on AI use in May 2026. Both make clear that AI-submissions are treated as the party’s own work and misleading evidence will be disregarded. In the UK, tribunal filings have risen sharply with judges reporting that a rise in AI drafted claims featuring fabricated precedents and excessive wordiness add to an already growing backlog of cases and increased costs for employers.
Practical takeaways for employers
It is therefore critical that employers:
- maintain thorough records and documentation;
- respond to claims using an itemised approach (where possible);
- expect polished, AI appearing claims and push for specifics i.e. dates, figures; and
- ensure HR and payroll systems can quickly and accurately produce auditable reports.
It is important to note that in this case the employee was not penalised for using AI itself but rather his failure to engage substantively with Ryanair’s defence. As clarified by the adjudication officer “a complainant is of course entitled to prepare his case with whatever assistance he chooses” but is not however “relieved of the burden of establishing that a deduction was made from wages properly payable to him”.
The clear message for employers is not that AI is prohibited from being used in employment proceedings, but rather that a well-documented, evidence-led response will exploit the gap in a generic AI drafted claim. This case also underlines that having a thorough pre-hearing response strategy (here, three detailed, consistent written breakdowns over seven months) can effectively set the terms of the evidential battlefield well before any hearing takes place.
If you would like to discuss any of the topics raised by this article, please contact Ailbhe Dennehy, Cliona Coleman or Leah Treacy in our Employment, Pensions and Benefits group, or your usual Matheson contact.
