Employment Tribunals and Generative AI – Access to Justice or Administrative Overload?

This article first appeared in the October 2026 issue of PLC Magazine (https://uk.practicallaw.thomsonreuters.com/Browse/Home/Resources/PLCMagazine).

Employment tribunals were designed to provide an accessible forum in which workers and employers can resolve workplace disputes without the formality and cost of ordinary civil litigation. Generative AI (GenAI) is now testing that model in an unexpected way.

GenAI technology can help an unrepresented claimant identify possible causes of action, structure an ET1 tribunal claim form, calculate loss and produce written submissions in minutes. While this may improve access to justice, it also reduces almost to zero the marginal cost of producing an extra allegation or extra 20 pages of argument. However, the cost of reading, responding to and adjudicating that material remains very human.

A system under pressure

This matters because the tribunal system is already under severe pressure. In 2025/26, employment tribunals received around 50,000 single claims and disposed of around 26,000. Single-claim receipts rose by 39% year on year, disposals fell by 12%, and the open single-claim case load reached 64,000 at the end of March 2026 - 55% higher than it was a year earlier.[1]

As at April 2026, the average time from a claim being filed to disposal was 59.6 weeks. While many one-day and two-day hearings were still being listed in 2026, some regions were listing longer hearings into 2028 and 2029.[2] While AI did not create this backlog, recent judicial guidance suggests that it may be increasing both the volume and complexity of work arriving at tribunal offices.

The interim relief warning

The clearest evidence of this trend comes from interim relief. Interim relief is available in a limited category of unfair dismissal claims, including certain whistleblowing cases (Section 128, Employment Rights Act 1996 (ERA)). An application must be made within seven days of the claimant’s effective date of termination. If successful, the tribunal may order reinstatement of re-engagement, or make a continuation of contract order that preserves the claimant’s pay and benefits pending the final hearing.

The threshold for interim relief is deliberately high. The claimant must show a “pretty good chance” of establishing the relevant case at final hearing. Most applications fail.

Historically, that scarcity was reflected in the number of applications. In June 2026, the Presidents of the Employment Tribunals in England and Wales and Scotland issued joint presidential guidance which said that tribunals across Great Britain had previously received around 20 interim relief applications a year, but that most offices are now receiving a similar number each month. The guidance states that the increase is largely in protected disclosure cases and that applications often indicate the use of AI.[3]

That is a striking change because interim relief is not simply another procedural application. It is an emergency jurisdiction. Hearings have to be listed quickly and other cases may be postponed in order to accommodate them. The Presidents expressly warned that the low success rate, increased documentation and consequential costs applications were adversely affecting the administration of justice.

The guidance therefore seeks to restore proportionality. Interim relief hearings will ordinarily be conducted by video, before an employment judge sitting alone, and last no more than three hours. Tribunals may impose limits on bundles, witness statements and written submissions. Where parties provide more material than can be read in the allocated time, they may be required to identify only the most important documents.

There is no objection in principle to the use of AI, but AI-generated submissions are described as often being too long and complex, including irrelevant material and failing to focus on key issues. Litigants remain responsible for ensuring that their submissions are concise, relevant and accurate.[4]

The economics of GenAI litigation

The interim relief problem illustrates a broader issue. GenAI changes the economics of legal drafting. Producing a 30-page submission once required substantial time, legal knowledge or professional fees. A litigant can now generate something of that length almost instantly.

That is not necessarily a bad thing. Employment law is complex and legal representation is expensive. Used carefully, AI can help individuals to understand terminology, organise events chronologically, identify documents and express a claim more clearly. For litigants in person, these are potentially significant benefits. However, AI does not naturally ration arguments. It can identify every theoretically available cause of action, product extensive authority lists and elaborate on weak points as readily as strong ones. A lay person may interpret length and legal vocabulary as evidence of quality.

The asymmetry is obvious. The claimant may incur almost no additional cost in producing an extra ten pages of arguments. The respondent employer must still read them, take instructions, assess whether an answer is required and potentially pay advisers to respond. A judge or legal officer must then identify what is relevant. AI-generated abundance can therefore impose real costs on every other participant in the system.

The risks are not limited to claimants. Employers and their advisers can use the same tools. An unfocused AI-assisted claim answered by an equally expansive AI-assisted response simply increases the volume of material without improving the tribunal’s ability to identify the issues.

Correlation is not causation

It is important, however, not to overstate the evidence. In March 2026, the government said that some stakeholders had reported potential claimants using AI to assess the strength of claims or help draft them, but that no formal assessment had been made of AI’s impact on the tribunal caseload.[5] The rise in claims therefore cannot properly be attributed to AI on the available evidence.

There are also structural explanations for the tribunal delays that long pre-date the widespread use of GenAI. These include reduced judicial and administrative budgets and capacity, the complexity of discrimination and whistleblowing cases, case-management demands and the after-effects of previous backlogs.

The more defensible proposition is narrower. AI is reducing the friction involved in initiating and pursuing litigation, and the judiciary now has direct evidence of that effect in at least one procedural area. It may also increase the amount of work that is associated with each case, which is not captured simply by counting the ET1s. That distinction matters. The future pressure on tribunals may arise not only from an increase in claims, but from each claim containing more allegations, applications, documents and submissions.

Reforms may increase demand further

The timing of these changes is significant because important statutory changes are about to broaden access to the tribunal system.[6] For most employment tribunal claims arising on or after 1 October 2026, the ordinary limitation period will increase from three months to six months (The Employment Tribunal (Extension of Time Limits) (Miscellaneous Amendments and Transitional Provisions) Regulation 2026 (SI 2026/758)). From January 2027, the qualifying period for ordinary unfair dismissal will fall from two years to six months and the statutory cap on compensatory awards for unfair dismissal will be removed (Sections 108 and 124, ERA, as amended by Section 25, Employment Rights Act 2025).

These reforms have policy objectives that are independent of AI. The longer limitation period gives parties more time to use internal processes and conciliation; the unfair dismissal reforms extend employment protection to a much larger number of employees. However, they will operate in a world in which identifying and articulating a potential claim has become materially easier. More people will have substantive rights, more time in which to pursue them, and inexpensive tools that can assist with drafting.

The government is increasing tribunal capacity. Funding has been provided for 32,590 employment tribunal sitting days in 2026/27, a virtual region allows remote hearings to be listed without geographical restriction and 34 additional salaried employment judges were due to begin sitting from September 2026. Recruitment is also under way for a further 55 salaried judges that are expected to add capacity from summer 2027.[7]

Whether these changes will be sufficient will depend not merely on claim numbers, but also how efficiently cases can be narrowed and managed.

What should employers do?

For employers, the immediate lesson is not to assume that an AI-assisted claim is either sophisticated or worthless. The quality will vary widely.

The more important response is to strengthen the underlying employment process. If an employee can feed a dismissal, grievance or whistleblowing dispute into an AI system and receive a plausible list of claims within minutes, weak documentation and inconsistent decision-making become easier to expose.

Contemporaneous records therefore matter more, not less, than before. Employers should be able to demonstrate why decisions were made, what evidence was considered, whether comparable cases were treated consistently and whether their own procedures were followed.

Whistleblowing dismissals require particular care because interim relief can create immediate procedural and financial consequences. The seven-day deadline means that an employer may receive a substantial application almost immediately after dismissal. HR and legal teams should know in advance how to escalate such an application and who is responsible for preserving the relevant documents and preparing the response.

Respondents should also resist the temptation to meet volume with volume. The guidance is a useful model for matters beyond interim relief. The key steps are to isolate the actual legal tests, identify the documents that matter and focus submissions accordingly. A strategy to respond effectively to an overlong AI-generated pleading may be disciplined definition of the issues rather than a line-by-line rebuttal of every argument.

A system problem requires a system response

There is no realistic prospect of banning AI from employment litigation, nor would that necessarily be desirable. The technology can improve access to legal information for people who cannot afford representation. The challenge is to capture that benefit without allowing automated drafting to force unlimited processing costs onto tribunals and opposing parties.

The guidance points towards one answer: stronger procedural discipline. Page limits, word limits, standardised structures, early issue identification and active case management can make the cost of assessing material more proportionate to the ease with which it can now be generated.

AI may also become part of the tribunal’s response. Subject to appropriate safeguards, technology could assist with chronology building, identifying duplicate allegations, organising evidence and highlighting the issues on which the parties actually disagree. HM Courts & Tribunals Service has already developed reasonable AI principles, and continues to modernise case management systems.[8]

The central issue is therefore not whether litigants should use AI; they already do. It is whether tribunal procedure can adapt quickly enough to a world in which the supply of legal text is effectively unlimited. Employment tribunals depend on accessibility, but accessibility is not the same as absence of discipline. If AI helps a claimant to present a genuine case clearly, it advances the purpose of the system. If it produces dozens of weak applications and hundreds of pages that judges cannot realistically read, it will delay justice for everyone else.

The task for tribunals, practitioners and employers is to preserve the first outcome while designing procedures that are robust enough to prevent the second. In employment litigation, the most valuable use of AI may ultimately not be producing more argument, but identifying those points that do not need to be argued at all.

[1] www.gov.uk/government/statistics/tribunals-statistics-quarterly-january-to-march-2026/tribunal-statistics-quarterly-january-to-march-2026

[2] https://questions-statements.parliament.uk/written-questions/detail/2026-07-01/15209

[3] www.judiciary.uk/wp-content/uploads/2026/06/PG-on-IRF-June-2026-final.pdf

[4] www.judiciary.uk/wp-content/uploads/2026/06/PG-on-IRF-June-2026-final.pdf

[5] https://questions-statements.parliament.uk/written-questions/detail/2026-02-24-h14836

[6] See PLC feature article “Employment Rights Act 2026: rights, rules and remedies”, www.practicallaw.com/w-049-2018

[7] https://questions-statements.parliament.uk/written-questions/detail/2026-07-01/15209

[8] https://questions-statements.parliament.uk/written-questions/detail/2026-02-24/h14836

-----

We hope you've found this article of interest.

If you'd like to discuss it, or any other matters where we may be able to assist you, please contact David Rankin on david.rankin@puntersouthallgroup.com.