AI in workplace disputes: Managing risk in a changing landscape

Employees are increasingly using tools such as ChatGPT and Copilot help with everyday tasks, including writing workplace grievances and letters, hoping to make them more comprehensive and persuasive. While AI may help employees to articulate issues, AI-generated grievances are often lengthy, highly legalistic complaints which may look impressive but can include irrelevant, duplicated or inaccurate content. 

This creates challenges for HR teams. Even where claims lack merit, each point still needs to be considered and addressed, increasing time and cost. AI-generated grievances can also entrench positions and make early resolution more difficult, particularly where AI-generated content creates unrealistic expectations about the strength or value of a case.

What can you do? 

• Define the scope early:  Identify the core issues at the outset. A short meeting with the employee can often clarify matters more quickly than working through a lengthy written complaint.
• Focus on the key points:  Address the material allegations properly but avoid being drawn into responding in detail to every peripheral issue.
• Strengthen internal processes:  Consider updating your grievance procedures to emphasise early informal resolution and to help avoid issues escalating.
• Educate employees on AI’s limits: Inform employees that while AI might help in drafting, it is not always legally accurate. Encourage employees to seek proper advice or to use internal support channels to avoid confusion.

AI is also impacting Employment Tribunal litigation, with rising claims and a growing backlog. There has been a structural shift toward “open track” claims (discrimination and whistleblowing cases) which are more complex and time consuming. The caseload is at record levels, with official statistics showing an increased and continuing rise in open cases. 

Employers are encountering longer ET1s (often with multiple causes of action), more frequent procedural applications (including interim relief) and issues around witness evidence credibility, particularly where AI may have been used to draft witness statements. Minutes of the 55th Employment Tribunal National User Group meeting in March 2026 noted the shared view of employment tribunal judges that AI is contributing to:

• more complex pleadings,
• a rise in applications for reconsideration and for interim relief (which had historically been rare and hard to win), and
• inflated schedules of loss and expanded claims.

It was also noted that Acas have begun to develop and test AI tools to streamline and improve their resolution services, such as how to best identify cases for triage.

What can you do? 

• Push early on for an agreed list of issues to narrow the scope of the case.
• Challenge unclear allegations and seek further particulars. Do not accept vague or overly broad allegations.
• Test witness evidence carefully to ensure that it reflects the witness’s own knowledge.
• Keep a clear audit trail of drafting and review processes, particularly where AI may have been used.

Growing adoption of AI in the workplace has led to working practices such as managers pasting investigation notes or draft outcomes into AI tools, HR teams using AI to sense-check grievance responses and employees inputting sensitive allegations and correspondence into AI tools. However, recent cases in the UK and the US have warned about the use of public AI tools (such as ChatGPT) in a legal context. 

In UK v Secretary of State for the Home Department (2026), known as the Hamid case, the Upper Tribunal made clear that uploading confidential or privileged material into an open AI tool (such as ChatGPT) is effectively placing it in the public domain and can breach confidentiality and waive legal professional privilege. Once privilege is lost, it is permanently waived and cannot be restored. 

The Tribunal drew a distinction between:

• Public/open AI tools (high risk of disclosure), and
• Closed, secure systems (lower risk, subject to controls).

In United States v Heppner (2025), the court found that material generated via an AI chatbot was not protected by attorney client privilege, partly because there was no reasonable expectation of confidentiality when using an open AI system. 

The cases highlight that using public (open) AI tools for dispute related work may waive privilege, breach confidentiality obligations, and create data protection risks (particularly where personal data is involved). Guidance for Judicial Office Holders on AI (last updated in October 2025) also warns that information entered into public chatbots may be retained and reused, and should be treated as potentially public. 

What can you do?

1.    Set clear boundaries on AI use to only allow the use of enterprise, closed AI systems where AI assistance is required.

2.    Train staff and managers on AI related confidentiality risks.

3.    Update policies and contracts to reflect AI specific confidentiality expectations (i.e. that no privileged or sensitive information should be input into public AI tools).