Picture a familiar scene: an online regulatory hearing in which a registrant faces allegations of professional misconduct. The panel chair asks a probing question to the registrant about their state of mind. There is a brief pause, and then a composed, well-structured answer follows. The answer is measured, coherent, and references a relevant regulatory standard. The panel is impressed by the registrant’s apparent clarity and self-awareness.
What the panel does not know is that the registrant has a second device open beside them, running an AI assistant that listened to the question, generated a suggested response within seconds, and displayed it on screen for the registrant to read aloud without anyone else knowing. This could even be done through a device within one’s ear, to make this seem even more natural, and that has been known to be deployed during in-person hearings.
This is not hypothetical speculation; these types of tools are freely available today and could be deployed across legal and regulatory proceedings throughout the United Kingdom. The question for regulators is no longer whether AI is present in their hearings. The question is whether they are aware of it and are prepared for it.
What do we mean by AI advocacy?
For the purposes of this article, “AI advocacy” refers broadly to the use of artificial intelligence tools to assist in the preparation, presentation, or conduct of a case before a regulatory body. The highest-risk category is real-time AI assistance during hearings where tools can listen to proceedings, transcribe and analyse what is being said, and generate suggested responses or arguments in real time for a party to deploy. The defining characteristic of this category is its immediacy and its invisibility. It occurs in the moment, without prior disclosure, and – when used effectively – can be extremely difficult to detect.
Why online hearings are particularly susceptible
Largely a product of the pandemic and lockdown, online hearings have delivered genuine and lasting benefits including accessibility, cost savings, and flexibility. But they have also created conditions in which AI assistance is uniquely difficult to detect and control. A participant appearing remotely may have multiple devices, numerous open browser tabs, and AI applications running simultaneously, entirely outside the view of the panel.
In fitness to practise and professional discipline proceedings, panels are frequently required to assess the credibility, insight, and remorse of a registrant. These assessments are often made on the basis of how the registrant presents themselves i.e., how they answer questions, and whether their account is consistent and coherent. If a registrant’s responses are being generated or substantially shaped by an AI tool in real time, the panel may be assessing the AI’s output rather than the registrant’s genuine understanding and reflection. Where a registrant gives oral evidence under oath or affirmation, the integrity of that evidence is fundamentally compromised if they are reading AI-generated answers from a second screen.
Confidentiality, data protection, and bias
Regulatory hearings routinely involve highly sensitive personal data which can include information relating to medical records, accounts of harm, financial information, and descriptions of deeply personal conduct. When a party uses an AI tool in real time, they are transmitting that data to a third-party system for processing. In many cases, this will include special category personal data within the meaning of the UK General Data Protection Regulation (UK GDPR). Parties may be entirely unaware that their use of AI tools constitutes a data protection breach.
The Judicial AI Guidance for Judicial Office Holders (October 2025) addresses this directly, requiring that confidential or sensitive information must not be entered into AI tools, and that judicial office holders must be alert to the risk that data entered into AI systems may be retained, used for training, or accessed by third parties. The same principle applies with equal force to parties in regulatory proceedings: inputting case-sensitive material into a consumer AI tool is not a neutral act.
Regulators should also be alert to the risk of bias. The Judicial Guidance notes that “AI tools based on LLMs generate responses based on the dataset they are trained upon” and that “information generated by AI will inevitably reflect errors and biases in its training data“. In a regulatory context, this could affect the framing of submissions, the characterisation of clinical or professional standards, and the presentation of evidence. This could happen without the panel being aware of the distortion.
The Law Commission’s Discussion Paper on ‘AI and the Law’ also raises questions regarding the issue of reliance on AI “where the persons using AI systems do not know the technical details underlying them”, and where there is risk of “undetected bias in the data” can lead “to harm that was not foreseen”. In a regulatory setting which has a public function, arguably the human decision-makers within a fitness to practise panel could unknowingly be relying on this output produced by an AI system but submitted by a registrant.
A further risk is the potential loss of confidentiality and legal professional privilege where AI tools are used to process case-sensitive material. In the case of R (on the application of Munir v Secretary of State for the Home Department, client letters and decision letters from the Home Office had been put into an open-source AI tool. It found that uploading confidential documents into public, open-source AI tool is equivalent to publishing them on the internet and breaches client confidentiality and waives legal professional privilege.
The AI Judicial Guidance reinforces this, directing that users “should treat all public AI tools as being capable of making public anything entered into them.” A registrant who inputs confidential case material – including accounts of alleged misconduct, prompting a consumer AI chatbot to assist with submissions, or other information which is relevant to their response – may unwittingly destroy the confidentiality of that material. Beyond that, opposing parties or regulators may seek disclosure of precisely how AI tools were used (or still being used during the proceedings), including the prompts entered, whether client or third-party information was processed, and whether any privilege has been waived as a result. Regulators should be alert to these risks and consider whether their procedural frameworks adequately address the disclosure obligations that may arise.
What can regulators do?
The risks are real and immediate, but they are not insurmountable. The Judicial AI Guidance establishes a clear framework for responsible AI use built around six core principles: understanding AI and its limitations; upholding confidentiality and privacy; verifying AI outputs; being aware of bias; taking personal responsibility; and being aware that court and tribunal users may themselves have used AI tools. Regulators should adopt an equivalent framework tailored to their own proceedings.
In practical terms, this means the following:
Develop clear policy and disclosure requirements. Regulatory bodies should amend their standard hearing directions to require parties to declare, in advance, whether AI tools have been used in the preparation of documents or submissions, and whether they intend to use AI tools during the hearing itself. This disclosure could be made by way of a signed declaration. The Judicial AI Guidance acknowledges that legal representatives bear professional responsibility for the material they put before a tribunal, and that AI use is “dependent upon context” as to whether it requires disclosure. For regulatory hearings, the use of AI during the hearing may warrant the need for mandatory disclosure.
Update hearing protocols. Participants should be required to confirm at the outset of a hearing that they are not using AI tools to generate or suggest responses in real time. Consideration should be given to requiring participants to close all applications on their devices other than the hearing platform during oral evidence.
Train panel members and legal advisers. The Judicial AI Guidance emphasises that those using AI must first ensure they “have a basic understanding of their capabilities and potential limitations”. Those conducting regulatory hearings should understand what AI tools are, how they can be used, and how to respond if AI use is disclosed or suspected of having been deployed; including adjourning, seeking clarification, or referring for further investigation.
Establish meaningful consequences. The primary safeguard against undisclosed AI use cannot be detection, and reliable detection is difficult. It must instead be a combination of clear rules, disclosure requirements, and meaningful consequences for non-compliance, supported by a culture of transparency that makes disclosure in advance the expected and standard course of action.
The integration of AI into everyday professional and personal life is accelerating rapidly. The judiciary has already recognised this challenge and has responded with substantive guidance. Regulatory bodies must now do the same. The Judicial Guidance makes clear that any use of AI must be “consistent with the overarching obligation to protect the integrity of the administration of justice”. That obligation is no less pressing in regulatory proceedings. The tools are already available. The vulnerabilities already exist. The time for regulators to act is now.
Sources:
- Artificial-Intelligence-AI-Guidance-for-Judicial-Office-Holders
- Law Commission – AI and the Law: Discussion Paper
About the author
Jac Davies is a Associate in our Regulatory team. He leads investigations into the conduct of regulated professionals on behalf of a range of regulatory bodies, including the Education Workforce Council (EWC) and the Teaching Regulation Agency (TRA).
