Generative AI in the Employment Tribunal – the rise of the machines

Anyone involved in litigation in the Employment Tribunals in the current environment is likely to observe a number of themes, including:

• the huge delays, which are getting longer. The most recent statistics from the Ministry of Justice, published on 11 June 2026, show that single Employment Tribunal claim receipts increased by 39% in 2025/26, while case disposals decreased by 12%. The open caseload also increased by 55% over the same period, as case receipts have exceeded disposals for the last few years. The backlog of claims is only likely to worsen from January 2027 with the removal of the compensation cap and the reduction in the period of qualifying service for unfair dismissal claims. Postponements of listed hearings (whether case management or final) are now increasingly common, with some cases having multiple postponements due to the lack of available judges;
• the extent to which claims are now being more commonly presented by litigants in person (LIPs). Government statistics from 2023 show that one in every three cases now has a litigant in person running their own case;
• the extent to which documentation from LIPs is increasingly voluminous and quotes at length what appears to be the law in the form of statutes and case law.

While part of the reason for the rise of LIPs in the Employment Tribunal is due to there being no legal aid in the tribunal system, it is fuelled to a considerable extent by the rise of claims prompted by generative large language model (LLM) artificial intelligence (GenAI) platforms like Gemini, ChatGPT and Claude, which are now routinely used to draft internal workplace grievances, early conciliation documents, and formal ET1 claim forms.

At one level, there is an argument that GenAI democratises access to justice for claimants who cannot afford legal representation, and thereby offsets the imbalance between them and employers who often have access to specialist employment law advice. Of itself, this appears to be in accordance with the Employment Tribunal Procedure Rules 2024, rule 3, the overriding objective, part of which considers whether the parties are on an equal footing.

At another level, however, because LIPs typically have little understanding of the way in which employment law operates and consequently cannot check GenAI output, that product is often presented unchecked, which leads to a variety of problems, as follows.

Hallucinations

The architectural design of GenAI tools is optimised for linguistic plausibility over factual accuracy, which can lead to hallucinated cases and fake citations as a result of GenAI fabricating convincing but fictional information. In tribunal submissions, this can manifest as non-existent case law, non-existent quotes, invented statutory provisions, and/or fabricated ACAS guidelines.

In one instance, Penningtons Manches Cooper acted against a LIP who presented voluminous documents to both the employer and the tribunal, including a document containing 46 cited cases. Quite remarkably, all 46 cases were inaccurate: nine were references to wholly fictitious cases, and the other 37 were misrepresentations of real ones. These hallucinations featured realistic case names, authentic-looking neutral citations, and cohesive summaries.

Significant time and administrative resources were required to search databases and prove the inaccuracies, and to prepare a citation table for the tribunal detailing the errors. As a result, the claimant doubled down on his GenAI use, and presented a skeleton arguing that his errors were ‘honest’ and ‘inadvertent’, and cited a further hallucinated case which he claimed recognised that difficulties, such as family medical emergencies affecting LIPs, could be relevant to the exercise of case management powers such as costs orders. That did not sit particularly well with the tribunal, and a costs order for several thousand pounds was made against the claimant.

Verbiage

Before the rise of GenAI, an unrepresented claimant’s grievance and/or ET1 form was typically a concise (often emotionally raw) short document detailing their core complaints. With the rise of GenAI, claimants can input a few bullet points into a prompt and receive dense, multi-page formal documents, often packed with complex legal terminology.

This creates a massive administrative burden. HR teams and defence solicitors will spend hours dissecting pages of text to locate the genuine, triable issues hidden beneath layers of AI-generated hyperbole. The case referred to above had vast multi-page submissions presented at various points (one of the documents was over 40 pages long). What is required in these sorts of cases is firm tribunal management on document submission, with the application of sanctions for breach (costs).

Over-affirmation and obstruction of settlements

LLMs are designed to be ‘agreeable assistants’. If an individual inputs their subjective narrative into an AI tool, the system will routinely validate their perspective. This can result in framing modest workplace misunderstandings as severe breaches of statutory rights, clear-cut acts of discrimination, or even a case likely to lead to the rarely-awarded interim relief (indeed, such is the extent of GenAI guidance resulting in interim relief applications, where near certainty of a final outcome is required, that there has recently been presidential guidance issued on applications for interim relief – see our previous article here).

This ‘over-affirmation’ instils an artificial sense of confidence in claimants whereby they believe they have ironclad, high-value cases. Often the result is that, because they have no independent verification, employees refuse reasonable early settlement offers via ACAS, and instead drive weak and/or legally incoherent claims to the tribunal. Of course, if a reasonable offer is properly framed on a ‘without prejudice save as to costs basis’, with the claimant being invited to seek independent legal advice, the possibility of a costs application can arise.

Data breaches

A further issue that arises is that if claimants enter their employer’s confidential information into a public source GenAI platform, there is a good chance they might be committing a data breach and/or a breach of confidentiality. While this will not assist after the event in relation to, for example, the reason for a dismissal in an unfair dismissal claim, it might assist in a breach of contract defence or with a Polkey-type reduction in capping loss for unfair dismissal.

Accordingly, while GenAI might superficially appear to assist claimants in the running of their cases, in the wrong hands (which can sometimes be lawyers’ hands – there have been a number of cases where lawyers and even the judiciary have been criticised for improper use of AI tools) it can be an example of little knowledge being a dangerous thing. There is a serious risk of more and more claimants receiving costs orders against them or their cases being struck out, all while continuing to grow the backlog of claims in the tribunal system.