Generative artificial intelligence is changing the way workplace disputes are presented. Employers increasingly receive grievances, appeals and tribunal documents that look polished and authoritative, yet mix legitimate concerns with repetition, irrelevant material, fabricated legal concepts and factual errors. The UK Employment Appeal Tribunal’s decision in Hancox v Sutherland and Others [2026] EAT 139 is therefore timely. Although not binding in Hong Kong, it offers a practical lesson: AI may assist drafting, but it does not displace human responsibility for accuracy, relevance and proportionality.
What happened in Hancox?
Mr Hancox worked for the National Farmers’ Union as a handyperson and was dismissed for misconduct during his probationary period. He brought claims against four individual managers, including whistleblowing detriment, automatic unfair dismissal and disability discrimination. In 2021, the Employment Tribunal struck out those claims because his scandalous, unreasonable and vexatious conduct — including a social media campaign — had made a fair trial impossible.
He appealed. Two days before the preliminary hearing, he filed a 300-page skeleton argument produced using ChatGPT, despite the applicable Practice Direction requiring between five and 15 pages. The required supplementary bundle had not been prepared.
The EAT accepted that generative AI may improve access to justice, particularly for unrepresented litigants. Its concern was not the use of AI itself, but unchecked output: inaccurate law, fabricated or distorted facts, excessive length, poor focus and non-compliance with procedural requirements.
The EAT made clear that every litigant — represented or not — remains responsible for documents filed in their name. AI-assisted material must comply with the rules, be checked against the factual, evidential and legal position, and address the central arguments without undue repetition. A party cannot submit raw AI output and leave the opposing party or tribunal to verify it.
Drawing on R (Ayinde) v London Borough of Haringey [2025] 1 WLR 5147 and published judicial guidance, the EAT applied the principles of personal responsibility and accuracy to litigants in person as well as regulated professionals.
Importantly, serious procedural failure did not extinguish a potentially valid point. With counsel’s assistance at the hearing, Mr Hancox advanced one arguable ground: whether he had received fair notice of the strike-out application and a fair opportunity to respond. Permission was granted for that ground to proceed to a full hearing.
The lesson for HR is important: even a grossly non-compliant document may contain a legally significant issue that must be identified and addressed.
Why the decision matters in Hong Kong
Hong Kong does not have the UK’s general statutory unfair dismissal regime. That distinction should not, however, encourage employers to treat dismissals — particularly during probation — as risk-free. Contractual notice, wages and statutory entitlements remain relevant, together with protections relating to discrimination, pregnancy, sickness, trade union activity and occupational safety. Depending on the facts, these protections may restrict or prohibit termination irrespective of length of service.
The decision may have even greater practical relevance in Hong Kong because legal representation is not permitted in the Labour Tribunal and parties are ordinarily required to conduct proceedings themselves. Employees and employers may therefore increasingly rely on generative AI to prepare claims, responses, witness statements and written submissions. While AI may help parties organise and present their cases, heavier reliance also increases the risk of excessive, inaccurate or legally misplaced material. The need for careful human review, procedural compliance and proportionate case management is accordingly particularly acute in Hong Kong.
What should HR do with an AI-assisted complaint?
1. Find the grievance inside the document
HR teams are already receiving AI-assisted grievances, discrimination complaints, termination appeals and Labour Tribunal documents. Their presentation may resemble formal pleadings or lawyer-drafted demands rather than workplace correspondence. Highly formatted text and legal language may nevertheless conceal repetition, irrelevance and ‘hallucinated’ facts or law.
A long or poorly drafted complaint should not simply be ignored. Buried within dozens of pages may be a serious allegation of discrimination, unpaid wages, retaliation, misuse of personal data, bullying or harassment. Equally, a line-by-line response may consume disproportionate resources and invite further correspondence. The better course is to acknowledge the complaint promptly, identify the material allegations and ask the employee to confirm the conduct complained of, the persons involved, the relevant dates, the supporting documents and the outcome sought. The employer can then settle a clear list of issues and investigate them under its usual procedure.
2. Keep responsibility with the author
Hancox does not suggest that AI-assisted communication is inherently improper. The difficulty was that the output was excessive, insufficiently filtered and presented in a way that shifted the burden of checking it to others. The relevant question is not whether AI was used, but whether a person has taken responsibility for the final product.
The same principle applies to employers. An employee’s complaint should not be rejected merely because AI was involved. Conversely, any AI-assisted disciplinary letter, investigation report, meeting note or termination communication must be checked carefully. A fabricated chronology, inaccurate dismissal reason or confident misstatement of Hong Kong law may become damaging evidence in later proceedings.
AI can help organise information and improve expression. It should not make factual findings, assess credibility, select the reason for dismissal or replace legal advice. Those remain human decisions for which the employer is accountable.
A practical protocol for Hong Kong employers
1. Update grievance and disciplinary procedures
Policies should allow the employer to request clarification, a concise summary and supporting documents where a submission is lengthy or unclear. Any page or format limits should be reasonable and flexible enough to accommodate disability or language-related needs. The employer should also make clear that its investigation and response will focus on the material issues identified.
2. Control the use of AI
Confidential employee or business information should not be entered into public AI tools without appropriate authority and safeguards. All factual assertions, legal references, quotations and calculations should be checked. Employers should also retain a clear record of who reviewed and approved the final document.
3. Keep the response proportionate
A sound outcome letter should identify the issues investigated, the evidence considered, the findings reached and any next step or appeal route. It need not answer every rhetorical point or repeat every piece of evidence. Concision is not avoidance where the employer has fairly addressed the substance.
The key takeaway
Hancox is not a warning against AI. It is a warning against unchecked AI. Courts and tribunals are likely to expect AI-assisted material to meet the same standards as any other document: accuracy, concision, relevance and compliance. Employers should expect more AI-assisted complaints, not fewer.
The challenge is to avoid being drawn into auditing machine-generated argument at the employer’s expense while still identifying and addressing genuine legal and factual issues. The answer is a disciplined process: acknowledge, clarify, distil, investigate, document and respond proportionately. Used responsibly, AI can improve communication. Used without oversight, it can magnify confusion, cost and risk.
If you require assistance or have any query related to Employment issues, please do not hesitate to contact our team.
This article is for information purposes only. Its contents do not constitute legal advice and readers should not regard this article as a substitute for detailed advice in individual instances.
