The "Inequality of Arms" Problem: AI-Assisted Complaints and Representations

One of the more immediate challenges facing regulators is not the use of AI by regulated professionals, but the increasing use of AI-generated advocacy by all participants in the regulatory process.
 

A number of regulators are reporting a marked increase in lengthy, sophisticated submissions generated using large language models. Complainants and registrants alike can now produce detailed legal arguments, analyse regulatory guidance, identify procedural points and generate extensive correspondence within minutes of receiving a response from a regulator. This pattern is now being seen across sectors, with public bodies and ombudsmen noting increases in both the volume and complexity of AI-assisted submissions.

At one level, this is a positive development. Used well, AI can improve access to justice by enabling individuals to articulate complaints more clearly, overcome literacy barriers and engage more effectively with formal processes. However, it also creates significant practical challenges.

Regulators cannot respond at the speed of a chatbot. They remain subject to public law duties to investigate fairly, assess evidence properly and reach decisions that are reasoned, proportionate and defensible. Those duties are unchanged by the use of AI. Every submission, however it is produced, must still be considered in accordance with the regulator’s statutory framework and procedural rules.

The result is a growing asymmetry. Complainants and registrants can generate multiple rounds of detailed representations almost instantaneously, while regulators must invest significant time and resource in reviewing, analysing and responding to those submissions. In an environment of increasing caseloads and constrained resources, this creates a new form of “inequality of arms” - not between the parties themselves, but between the parties and the regulator responsible for resolving the matter fairly.

Quantity versus substance
 

A central difficulty is that AI can create the appearance of legal sophistication without necessarily adding evidential value.

Regulators are increasingly encountering submissions that are lengthy, highly structured and expressed in legal or regulatory language, yet add little by way of new evidence or meaningful analysis. In some cases, AI-generated material may include misquoted legislation, fabricated authorities or inaccurate legal assertions, all of which require time-consuming verification. Ombudsman services have similarly identified AI-assisted complaints that are excessively long, unfocused, or rely on incorrect material, slowing down case resolution[1].

There is therefore a real risk that volume and apparent authority obscure the true issues in dispute.

The appropriate response is a disciplined approach to case handling. Regulators are not required to engage with every point in an expansive submission. Instead, decision-makers should:

• identify the core issues in dispute
• focus on relevant evidence
• and address the matters that are material to the decision

Put simply, the task is to ensure that decisions are driven by substance, not overwhelmed by volume.

Managing AI-assisted submissions within existing frameworks
 

The most important point for regulators is that AI does not require new legal powers or a new framework. Existing procedural tools are generally sufficient - but they need to be used more actively.

Emerging guidance from the Information Commissioner’s Office[2] is helpful and readily applicable across the regulatory landscape: regulators can, and should:

• seek clarification where submissions are unclear or overly complex
• limit or narrow issues where the scope is disproportionately excessive
• rely on existing safeguards against repetitive, burdensome or vexatious material

In practical terms, this means controlling the scope of the correspondence / dispute at an early stage, rather than attempting to respond to everything that has been generated.

Targeted procedural controls
 

In addition, regulators may wish to consider whether targeted procedural adjustments are needed to keep their processes workable in an AI-enabled environment.

Where volumes and complexity begin to impact effective decision-making, proportionate measures may include:

• introducing structured formats or templates to focus submissions
• applying reasonable word or page limits
• issuing clear guidance encouraging concise, relevant and accurate representations

These are not about restricting access to the process. Their purpose is to ensure that issues can be properly identified, understood and determined, without being obscured by excessive or unfocused material.

There is no single model that will work across all regulators. Any changes must be proportionate, transparent and consistent with public law duties. But the direction of travel is clear: regulators should feel confident in designing processes that prioritise clarity and focus over volume.

Reassessing evidence: form versus reliability
 

A more subtle but important shift is that AI weakens the traditional link between how a submission looks and what AI also disrupts a more subtle assumption in decision-making: that the quality of written advocacy reflects the quality of the underlying case.

Historically, clarity, structure and legal coherence have often been treated as indicators of careful thought and reliability. That link is now weaker. AI can produce submissions that are polished, confident and legally framed, without reflecting a corresponding depth of understanding.

This has important consequences. Regulators may need to:

• place less weight on the form and style of written representations
• focus more closely on evidence, consistency and factual support
• test assertions rather than accepting them at face value

In practice, this will often mean relying more heavily on:

• contemporaneous documentation
• objective evidence
• direct clarification, including oral engagement where appropriate

The focus must remain on what is proved, not how it is presented.

Looking ahead: Can AI demonstrate insight?
 

These issues become even more acute in professional disciplinary contexts, where a registrant’s written reflections are often relied upon as evidence of insight, remediation and understanding.

AI can generate convincing, well-structured reflective statements which appear persuasive but may not reflect the registrant’s own thinking. This raises important questions about authenticity and evidential weight.

We explore this issue in more detail in our forthcoming blog: Beyond the Algorithm: Assessing Genuine Insight When Submissions Are AI-Generated.

A recalibrated approach
 

The challenge for regulators is not to resist AI, but to respond to it in a way that preserves fairness and decision quality.

A coherent approach should include:

• applying existing legal frameworks consistently, regardless of AI use
• actively managing scope and relevance in submissions
• prioritising substance over presentation
• reweighting evidence, particularly in relation to insight and remediation
• ensuring staff and decision-makers are equipped to engage critically with AI-generated material

At the same time, regulators should recognise the benefits. AI can improve access to complaint mechanisms and enable clearer articulation of concerns. Its use will only increase.

The objective, therefore, is not to slow AI down, but to ensure that human decision-making remains focused, proportionate and robust in a system where advocacy is no longer constrained by effort or resource.

Key takeaways for regulators
 

1. You must respond - but not to everything
   • Regulators remain under a duty to consider and respond to representations. However, you are not required to address every point raised. Focus your response on what is relevant to the decision.
2. Define and maintain focus on the real issue
   • Ensure the process stays anchored to the issues that matter. Lengthy or complex submissions should not distract from the core questions to be determined.
3. Focus on evidence, not presentation.
   • Well-drafted or legalistic submissions are not inherently persuasive. Decisions must be grounded in evidence.
4. Prepare for the next challenge: insight
   • AI-generated material raises difficult questions about authenticity and reflection - particularly in disciplinary contexts.