Something has changed in the inbox of every HR manager and employment lawyer in the country. The grievances landing on desks today are longer, more polished and more legalistic than they used to be. They cite case law. They quote the Acas Code. They reference statutory provisions and compensation figures. And, increasingly, they were not written by the employee at all — at least not in the conventional sense. They were drafted, structured or "improved" by a generative AI tool such as ChatGPT, Microsoft Copilot or Google Gemini.
Employment lawyers are now reporting a clear and accelerating trend of employees using AI to draft grievances, disciplinary responses, subject access requests and full employment tribunal claims. Some practitioners have even coined a nickname for the phenomenon: the "GIT" — a grievance invented by technology.
For employers, this is not a passing curiosity. It changes how complaints look, how long they take to resolve, what they cost, and how litigation plays out. Used well, AI can help an articulate employee express a genuine concern clearly. Used badly, it can inflate expectations, introduce fabricated law, breach data protection rules, and clog an already overburdened tribunal system. This article sets out what is happening, why it matters, the real cases that show how courts and tribunals are responding, and a practical playbook for employers who want to handle AI-assisted complaints fairly and defensibly.
The use of AI in employment disputes is no longer anecdotal. Ministry of Justice figures showed a 32 per cent rise in open tribunal cases in 2024–25, and with the Employment Rights Bill predicted to push case volumes up by a further 15 per cent, lawyers fear an influx of AI-generated claims will deepen the strain on the system.
Ailie Murray, an employment partner at Travers Smith, says her team is seeing more employees use AI to draft claims, while Jess Kelleher, head of litigation at Halborns, has received AI-assisted submissions citing false or irrelevant case law. The International Bar Association has described how HR teams and lawyers now struggle with the sheer speed at which litigants respond — lengthy correspondence turned around in a matter of hours — eroding goodwill at exactly the point where settlement might otherwise be possible.
The driver is simple. Generative AI gives an unrepresented employee something they have never had before: instant, confident, free "legal advice" and the ability to produce a document that looks professionally drafted. For litigants in person who cannot afford a solicitor, an AI chatbot may be their only source of legal assistance.
There is no foolproof test, and employers should never treat suspected AI use as a reason in itself to dismiss a complaint. But practitioners have identified a consistent set of tell-tale signs. Taken individually they prove nothing; taken together they can help you form a view.
Crucially, where it is not obvious, there is no problem in simply asking the employee whether their submission was created with the assistance of AI.
The single biggest risk in AI-generated legal material is "hallucination" — where the tool produces confident, plausible-sounding text that is simply untrue. Large language models predict likely word combinations from vast datasets; they do not verify accuracy against authoritative sources, and they can fabricate cases, citations, quotations and even legislation that does not exist.
This is not theoretical. In the employment context, lawyers report receiving submissions quoting cases that do not exist, applying US law to UK disputes, and citing imaginary Acas Codes. One widely reported example involved a claimant who relied on a fabricated authority, "Johnson v British Airways 2019" — a case ChatGPT invented wholesale — while another claimant working for a cheese company used ChatGPT to draft a tribunal claim that produced such poor-quality, fabricated legal principles it undermined what might have been a valid case.
The fundamental danger for the employee is that they remain personally responsible for everything they put before a tribunal, even if AI wrote it. As experienced tribunal users put it bluntly: if you assert that the law says one thing when it says another, you are misleading the tribunal — not the AI — and there is no defence in blaming ChatGPT. In the most serious cases, reliance on AI without verification could expose a litigant to a finding of contempt of court.
The defining UK authority is the High Court's June 2025 judgment in R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383, heard together under the court's "Hamid" jurisdiction for potential abuses of process. Both cases involved submissions containing multiple fictitious or materially inaccurate case citations, suspected to have been generated by AI and relied upon without verification.
Dame Victoria Sharp set out a series of stark warnings:
The court was explicit about the consequences. Deliberately placing false material before a court with intent to interfere with the administration of justice can amount to perverting the course of justice — a common law offence carrying a maximum sentence of life imprisonment — and, short of that, may amount to contempt of court depending on the person's state of knowledge. The court's available powers include public admonition, costs orders, striking out, regulatory referral, contempt proceedings and referral to the police.
By late 2025, commentators were tracking a steady stream of UK hallucinated-citation cases — into the twenties and beyond — repeatedly referencing Ayinde as the benchmark for the checks that should have been done.
The UK is not alone. The problem traces back to the US case of Mata v Avianca Inc (S.D.N.Y. 2023), where lawyers relied on AI-generated non-existent cases. In Noland v Land of the Free, LP (2025), California's first published appellate opinion on AI misuse, the court found 21 of 23 case quotations in the opening brief were fabricated and imposed a $10,000 sanction on counsel.
Two recent Gulf employment cases are especially instructive because they arose directly out of employment disputes:
The judiciary's position in England and Wales is permissive but firm. The updated Artificial Intelligence: Guidance for Judicial Office Holders (latest version October 2025) does not ban AI. It is built on three simple messages: generative AI predicts likely word combinations without verifying accuracy; confidential data should never be entered into public LLMs; and the user remains fully responsible for the output.
For litigants in person, the guidance tells judges it is appropriate to enquire whether AI was used, ask what accuracy checks (if any) were performed, and remind the litigant that they remain responsible for all material submitted.
Importantly for employers, tribunals have been pragmatic about the legitimate use of AI to help an unrepresented person express themselves. In Miss E Kaloudi Tsikni v Kontis & Alphakon Limited, the respondents argued that it was impermissible for the claimant to use ChatGPT to draft her witness statement. The tribunal was aware she had used AI tools and took that into account when weighing the evidence, but — having recalled her to give oral evidence in Greek through a translator, allowing her account to be properly tested — refused the application for reconsideration. As the commentary puts it: "There appears to be nothing to suggest that a tribunal will find fault with a litigant in person making use of AI tools to draft their evidence," provided the facts are true to the best of the witness's knowledge and belief; how those facts are put onto the page "appears to be of little interest to the Employment Tribunals".
The lesson for employers is clear: the mere fact that AI helped draft a grievance, response or witness statement is not a basis to disregard it. What matters is the truth and substance of what is said.
There is a second, often-overlooked risk that cuts the other way — against the employee, and potentially the employer. To get useful output, employees frequently paste sensitive material into public AI tools: meeting notes, disciplinary and grievance records, colleagues' names, even HR documents. Once that information is entered into a public platform, the employee "loses control of it", creating GDPR and confidentiality risks for the business.
For the employee, this can be a serious own goal. Uploading confidential information — including notes of meetings, disciplinary or grievance hearings, or employee records — to an AI tool may amount to unauthorised disclosure of confidential information, which is likely to be a gross misconduct offence potentially justifying dismissal.
For the employer, it raises incident-response questions. If sensitive personal data has been exposed, you may face notification obligations and should move quickly to understand what data was uploaded, to which platform, and what containment is possible — adjusting account settings, deleting interaction history, or contacting the provider. The judiciary's own guidance now signposts where to report inadvertent disclosures as data incidents — a useful reminder that this is a recognised category of risk.
Perhaps the most practically damaging effect of AI for employers is its impact on settlement. Joanne Frew, global head of employment and pensions at DWF, reports cases where AI told claimants they would receive unrealistically high compensation awards, "making it all but impossible to have a sensible settlement discussion".
The International Bar Association warns that AI can leave claimants entrenched and less willing to accept what would once have been a reasonable commercial offer, increasing the likelihood that cases run all the way to a final hearing. Employers may need to:
There is, however, a flip side that favours well-advised employers. AI-polished pleadings often mask foundational legal and factual gaps. Errors and fabricated citations present an opportunity to challenge the claimant's credibility and correct the legal foundation of the claim. A claim that looks formidable on the page may be vulnerable to early strike-out or dismissal applications, and a well-documented record of AI errors can support costs arguments where the conduct has been unreasonable.
Drawing the threads together, here is how employers and HR teams should handle AI-assisted grievances, disciplinaries and claims.
1. Treat the complaint on its substance, not its format
A grievance remains valid even if drafted with AI, and organisations cannot afford to treat AI-drafted grievances differently in principle from traditional ones. The underlying concern may be entirely genuine. Respond as you would to any complaint — with professionalism, empathy, and a fair process.
2. Always follow the Acas Code
Under the Acas Code of Practice on disciplinary and grievance procedures, you must investigate thoroughly and keep clear written records. The procedure you follow will be scrutinised if the case reaches a tribunal. AI does not change this baseline.
3. Get the employee talking — in person
The most effective response to a long, formalised, AI-assisted submission is a face-to-face meeting. Ask the employee to explain, in their own words, what the issues are. You can then focus your response on the key concerns confirmed verbally rather than addressing every line of a lengthy document. In-person discussion is consistently identified as the only reliable way to understand and resolve the real issue, and it naturally discourages purely AI-driven correspondence.
4. Refocus and triage
Where a submission contains irrelevant or off-topic material, HR teams should feel empowered to refocus the discussion with specific questions, identify the actual complaint, and apply the correct procedure. For accommodation or wellbeing requests that read as vague and generic, ask targeted follow-up questions to draw out the specifics AI stripped away.
5. Verify any law the employee cites
Do not assume cited authorities are real. Check case names and quotations against authoritative sources such as gov.uk, BAILII and Acas. Fabricated or misapplied law should be corrected on the record, both to keep the process fair and to protect your position in any later proceedings.
6. Put it in policy
Set clear expectations in your AI, grievance and disciplinary policies:
7. Train managers and HR
Equip front-line managers to deal with issues early, and train HR to recognise material details buried in overly formalised or legally complex submissions and to apply proportionate, fair procedures.
8. Reset litigation strategy
For claims, brief your advisers to flag AI indicators and verify citations, build a contingency into defence budgets, evaluate inflated demands promptly, and use early dispositive applications where the facts support them — AI cannot rescue a claim that is foundationally weak.
Conclusion
AI is now a permanent feature of the employment dispute landscape. It is empowering employees to raise concerns in more polished, structured and confident ways — sometimes helpfully, sometimes with fabricated law, inflated expectations and confidentiality breaches that damage their own position. The courts have made the governing principle unmistakable: AI is permitted, but the human user remains wholly responsible for accuracy and confidentiality, and fabricated material can carry consequences as serious as costs orders, contempt, or worse.
For employers, the response is not to fear the technology or to dismiss AI-assisted complaints, but to stay disciplined: follow the Acas Code, treat every complaint on its substance, get the employee talking in person, verify any law cited, protect confidential data, and set clear expectations in policy. Handled this way, the rise of the AI-drafted grievance becomes not an existential threat but a manageable — and in some respects opportunity-rich — evolution of an area employers already know how to navigate fairly.
Employment Tribunal Claims and How to Respond
Disciplinary & Dismissal Procedures