In July 2026, the Financial Conduct Authority issued the Mills Review examining the implications of advanced AI on consumers, retail financial markets and regulators. As part of the Review, it acknowledged that AI is already reshaping how complaints are handled across financial services, including the Financial Ombudsman Service. The restructuring and insolvency sector is no exception: insolvency practitioners (IPs) are increasingly encountering complaints that appear to have been generated, in whole or in part, by AI tools as noted by a recent Institute of Chartered Accountants in England and Wales' (the ICAEW) report entitled 'AI and Complaints - keeping the process human, proportionate and professional'.
As noted by the Mills Review, used carefully, AI can be a valuable tool for individuals to draft complaints. In practice, however, AI is frequently misused; individuals can provide a brief description to an AI tool, copy the output verbatim, and submit it without any critical review. The results are often complaints that are poorly targeted, factually unreliable, and padded with irrelevant legal arguments including, in some cases, entirely fictitious case law, legislation, or regulations. A single throwaway sentence can generate a submission running to twenty pages or more. As this pattern becomes entrenched, the volume of unfocused and low-quality complaints is likely to increase substantially.
A recent R3 report (the R3 Report) found that only 11% of survey respondents agreed that current regulations adequately address AI use in the profession. That gap extends beyond AI adoption to how IPs should handle the AI-generated correspondence and complaints already arriving on their desks. The implications for IPs are significant, and it is important to put measures in place to address the issue now. Whilst the ICAEW has published a recent report on AI and complaints, which offers some useful practical guidance for IP firms navigating AI-related complaints, the ICAEW's Code of Ethics 2025 (the Code) does not yet treat AI-generated communications as a discrete or standalone topic.
In this article, we consider where AI-generated complaints sit within the current IP regulatory framework and set out the practical steps IPs can take now to adapt their complaints-handling processes accordingly.
The regulatory landscape
The ICAEW's own 'Make a Complaint' webpage offers a telling indicator of how seriously the profession is beginning to take this issue. It warns individuals against using AI to prepare a complaint, cautioning that whilst AI tools can be helpful "…they also get things wrong, including referring to irrelevant or non-existing sources or adding arguments that may not be accurate or relevant. This can result in delays as we may need to clarify key points that have become buried in repetitive AI-generated wording before we can begin to assess the case." This is precisely the pattern the Mills Review identified: AI is lowering the quality of complaints and increasing operational strain on those responsible for handling them. That the ICAEW is now compelled to address this issue on its public-facing complaints page speaks to how quickly the problem has embedded itself in practice.
Yet despite this warning, as noted above, the ICAEW Code does not itself treat AI-generated correspondence as a discrete topic. It does, however, contain directly relevant guidance on the use of technology, bias, professional judgement, and the conceptual framework, all of which bear squarely on how an IP should approach a complaint that appears to have been generated by AI.
Rule 2105.1 – Professional behaviour
Regardless of the origin of a complaint, the principle of professional behaviour requires compliance with relevant laws and regulations, behaviour consistent with the profession's responsibility to act in the public interest in all professional activities and business relationships, and avoidance of any conduct that the IP knows or should know might discredit the profession.
An IP who responds to an AI-generated complaint by dismissing it as inherently unreliable, without engaging properly with its substantive content, risks being seen as avoiding legitimate scrutiny, which could itself constitute conduct discreditable to the profession. Therefore, when dealing with an AI-generated complaint, IPs should still engage with the complaints process transparently and in a manner that upholds the public interest and the reputation of the profession.
Rule 2111.5 (a) – Apply an inquiring mind
The Code requires insolvency practitioners to approach all professional activities with an inquiring mind, considering the source, relevance, and sufficiency of information, including whether it may be affected by bias, whether relevant information might be missing, and whether it provides a reasonable basis for any conclusions reached.
This applies to IPs when considering an AI-generated complaint. AI is known to produce plausible but factually inaccurate output, and an IP receiving such a complaint should scrutinise its underlying content rigorously: the factual assertions made, any case references cited, the statutory provisions relied upon, and any calculations or chronologies. An incoherent or obviously AI generated complaint is not, in itself, a basis for not accepting the complaint.
Rule 2111.5 (b) – Exercise professional judgement
The Code requires IPs to exercise professional judgement when applying the conceptual framework. This means applying relevant training, knowledge, skill, and experience to the facts and circumstances at hand, having regard to the nature and scope of the activities concerned and the interests and relationships involved.
Applied to AI-generated complaints, this obligation is relevant in two ways. First, an IP should not allow the potential superficial sophistication or complexity of an AI-generated complaint to stand in for substantive merit. Second, the technical complexity of a complaint does not diminish simply because it was AI-generated and/or because the complainant may not understand its substance. Rule 2111.5 (b) recognises that in such cases, professional judgement may still require the IP to consider whether their own expertise is sufficient or whether they should seek input from a suitably experienced colleague or adviser before preparing a response regardless of whether the complaint is AI-generated.
Rule 2115.1 A6 – Identifying threats associated with the use of technology
The Code sets out specific guidance on threats arising from the use of technology. Whilst this guidance is primarily directed at an IP's own use of technology, its conceptual framework is equally useful when evaluating technology used by a third party to generate a complaint.
The Code identifies the following as potential self-interest threats arising from technology use: The data available might not be sufficient for the effective use of the technology; the technology might not be appropriate for the purpose for which it is to be used; and the insolvency practitioner might not have sufficient information and expertise, or access to an expert with sufficient understanding, to use and explain the technology and its appropriateness for the purpose intended.
Therefore, an IP should consider the following when reviewing an AI-generated complaint and reflecting on how best to respond to it:
- Whether the AI tool used by the complainant was given accurate, complete, and contextually appropriate data about the conduct of the appointment, or whether there are incorrect assumptions or facts relied upon that must be addressed;
- Whether the AI tool has in fact resulted in a regulatory complaint about insolvency conduct; and
- Whether the complainant, or those advising the complainant, had sufficient expertise to verify and stand behind the AI's output before submitting it, or whether there are incorrect or hallucinated authorities or principles in the complaint which should be recorded in any reply.
Rule 2117.3 – Bias
Among the most directly relevant provisions in the Code is the risk of automation bias, defined as a tendency to favour output generated from automated systems, even where human reasoning or contradictory information raises questions about its reliability.
This risk cuts in both directions – when assessing an AI-generated complaint, an IP might (i) err by over-accepting the complaint's content on the implicit assumption that AI-generated output is objective or comprehensive, or (ii) fail to challenge specific allegations simply because they have been articulated in a sophisticated and apparently well-founded manner.
The Code equally flags confirmation bias, a tendency to give greater weight to information that corroborates an existing belief than to information that contradicts it. This could be relevant where the complaint is clearly AI generated with some information that might be hallucinated – this does not automatically mean there is nothing of substance to consider.
To mitigate these risks, the Code already recommends seeking expert input, consulting others to ensure appropriate challenge, and undertaking training on the identification of bias as part of continuing professional development.
Adapting your response process
Understanding the regulatory risks is one thing; building a process that guards against them is another. So, what steps can an IP take now to adapt their complaints-handling process and ensure continued compliance with the Code and their regulatory obligations?
Triage and documentation. When a complaint arrives, record not just its content but its apparent basis and the context in which it has been received. Where indicators of AI are present, note them in your file, not to dismiss the complaint, but to frame the internal response appropriately and to provide context if the matter escalates. This is consistent with the Code's requirement to apply an inquiring mind and to be conscious of automation bias, considering the source and sufficiency of the information before reaching any conclusions.
Respond to the facts, not the framework. The most effective counter to a legally sophisticated but factually thin complaint is a response grounded in the specific facts of the appointment. An AI tool can generate a complaint that correctly states what SIP 9 requires; it cannot know what decisions you made, when, on what basis, and with what creditor engagement. Your response should make that clear. A factually dense, appointment-specific reply will expose the hollowness of a generic complaint far more effectively than a mirror-image legal rebuttal, and will demonstrate the kind of substantive professional engagement and judgement that the Code requires.
Do not substantively respond with AI. When hit with an apparently unmeritorious, obviously AI generated complaint, there is a natural temptation to respond in kind. However, a generic, formulaic response to a complaint, even an apparently unmeritorious one, carries its own regulatory risk. The Code's professional behaviour principle requires conduct that upholds the public interest and the reputation of the profession; an AI response in kind is unlikely to satisfy that standard, and the regulators will be looking for evidence of genuine engagement with the allegation.
AI still has its uses. That is not to say that AI has no role in the complaints-handling process. It can potentially assist with summarising a lengthy communication or structuring a response, reducing administrative time without compromising the quality of the substantive reply. The R3 Report found that 52% of practitioners are already using generative AI tools such as Copilot and ChatGPT, typically on an ad hoc basis. Used selectively and with appropriate scrutiny, this is entirely consistent with the Code, but the Code's warning on automation bias is a reminder not to outsource professional judgement to these tools. The R3 Report also highlights that IPs should ensure they are using enterprise-grade platforms rather than consumer-facing versions (i.e. public versions of ChatGPT) when handling any client-related material, including complaints.
Your file is your best defence. None of this changes the fundamental position that the best protection against any complaint, AI-generated or otherwise, is a well-maintained file. Comprehensive records of decisions, creditor communications, remuneration approvals, and case progression remain the IPs' primary shield, and are the clearest demonstration that the Code's requirements have been met in practice.
Conclusion
AI-generated complaints are not a future risk; they are the present reality, and their volume is only likely to increase. The regulatory obligations for insolvency practitioners have not changed, but the environment in which those obligations must be discharged has shifted materially, and practitioners who fail to adapt their response processes risk falling short of the standards the Code demands.
The good news is that the Code already provides a coherent framework for navigating this landscape. The principles of professional behaviour, an inquiring mind, professional judgement, and an awareness of bias together equip practitioners with the tools they need, provided those tools are applied with genuine rigour.
On a practical level, the fundamentals remain unchanged: clear decision-making, thorough record-keeping, proactive creditor communication, and prompt, substantive engagement with any complaint received. These have always been the hallmarks of good insolvency practice. In this new complaints reality, they are more important than ever.
