AI in practice: When efficiency undermines judgement

Artificial intelligence (AI) is rapidly finding its place within professional services, offering the promise of greater efficiency, enhanced research capability and improved client delivery. However, a recently published judgment from the English High Court provides a stark reminder that these benefits come with material risks, particularly where reliance on AI displaces professional judgement. The judgment is a clear warning for professionals: technology may support your work, but it doesn’t reduce responsibility for accuracy, integrity or ethical decision making.

A routine application with unexpected consequences

At its core, the case of Anthony Malcolm Cork & Anor v Mark Smith [2026] EWHC 1199 (Ch) concerned a block transfer application under the Insolvency (England and Wales) Rules 2016 (IR).

Nothing remarkable: a well-established and largely administrative process that enables an outgoing officeholder to transfer multiple appointments to a replacement in a single application, typically determined without a hearing.

In this instance, the application asked for the transfer of appointments and the “release” of the outgoing officeholder from liability. The rules are clear for administrators, but the law is less clear and more uncertain when it comes to giving release to liquidators.

This led the court to ask a straightforward question: what statutory provision allowed a release clause to be included?

Taken together with the original application, the response prompted a wider review of professional standards in the age of AI.

The introduction of misleading material

In response to the court’s query, the applicant’s solicitors sent a letter saying there was a clear power to grant a release under IR 12.37(5). They even included a detailed quote of what they said was the relevant wording. However, that wording doesn’t appear anywhere in the IRs.

Judge Mullen quickly identified that the provision had been misquoted. This raised immediate concerns not only about the specific submission, but about the reliability of the material being placed before the court.

Those concerns were compounded when a second letter was submitted following Judge Mullen’s request for an explanation. Rather than clearly withdrawing the incorrect assertion, the correspondence attempted to explain the wording as a “summary” of various provisions within the IRs. The court found this explanation unconvincing and, in substance, misleading.

The role of AI: Plausible but incorrect

The reason became clear later: the use of AI. The judgment explains in detail how AI was used in this case, drawing directly on the chat transcript between the user and the AI model.
Evidence before the court revealed that a junior solicitor had used an AI tool to assist with researching the issue and drafting the response. The AI tool generated a coherent and authoritative-sounding passage that purported to set out the relevant law. In reality, it was entirely fabricated.

Crucially, the AI tool also issued explicit warnings that its output should be verified against primary legislative sources before being relied upon. Those warnings weren’t followed.

Instead, the AI-generated wording was incorporated into correspondence with the court and presented as an authoritative statement of the law. When challenged by the court, the same process was repeated. Further engagement with the AI tool was used to construct an explanation, rather than to investigate and correct the underlying error.

A failure of professional judgement

While the technology played a central role, the judgment makes clear that the issue wasn’t AI itself, but the way in which it was used.

Several failures of professional judgement can be identified:

1. Over-reliance on AI as a research tool

The junior solicitor appears to have treated AI output as a reliable source of legal authority, rather than as a starting point for further inquiry. This reflects a misunderstanding of the nature of generative AI, which produces text based on patterns rather than verified legal sources. There’s perhaps a greater risk of hallucination in niche areas where there’s less material for generative AI to work from.

2. Failure to verify against authoritative sources

At multiple stages, the opportunity to check the relevant statutory provisions was not taken. The court noted that even a basic review of the legislation would have revealed that the quoted wording was incorrect.

3. Lack of critical scepticism

Despite clear warning signs, including the AI’s own acknowledgement of uncertainty, the output was accepted at face value. Professional scepticism, a cornerstone of both legal and accountancy practice, wasn’t applied.

4. Inadequate supervision

More senior colleagues approved the correspondence without adequately checking the validity of the legal proposition on which it relied. The judgment reinforces that responsibility for delegated work remains with those supervising it, regardless of whether AI is involved.

5. Poor judgement in addressing error

Perhaps most significantly, when the issue was identified by the court, the response wasn’t to investigate and correct the position transparently. Instead, further reliance was placed on AI to provide an explanation. This not only failed to resolve the issue but exacerbated the court’s concern.

Ethical duties in a technological environment

The judgment didn’t go into the detail, but it seems that the relevant output may have been produced during an initial test or implementation phase of the firm’s use of AI.

The firm’s compliance officer and the chair of the firm’s Innovation and Change Board provided statements to the court setting out the firm’s AI policy and explaining its AI pilot. The court was advised of further steps that had since been put in place as guardrails on the use of AI.

This serves as a useful reminder from the ICAS Code of Ethics that appropriate safeguards must be put in place when using technology.

The judgment reaffirms that fundamental professional duties remain unchanged, irrespective of the tools used.

In particular, the court highlighted, among others, the duty to:

• Provide a competent and diligent service.
• Maintain transparency and candour, especially when errors occur.

AI risk: Understanding “hallucination”

A central concept in the judgment is the phenomenon of AI hallucination, the generation of information that is plausible in form but incorrect in substance.

The case demonstrates that:

• AI can fabricate statutory provisions and case law.
• AI can present such material with a high degree of confidence.
• AI may provide internally consistent reasoning, making errors less obvious.
• AI can even generate explanations of its own incorrect output, compounding risk.

These characteristics mean AI can be particularly risky in technical and regulated settings, where accuracy is most important.

Regulatory and disciplinary implications

Although the court accepted that there was no deliberate intention to mislead, it found that the correspondence amounted to misleading the court on two occasions. The matter had already been referred to the Solicitors Regulation Authority (SRA), which the court considered an appropriate course of action.

Importantly, the judgment notes that a different outcome may have followed had the error been identified early and addressed candidly. The failure to do so led to increased court time, additional cost and reputational consequences.

For regulated professionals, this underlines a key point: errors aren’t necessarily fatal, but how they’re handled is critical.

Practical lessons for accountancy professionals

While arising from a legal context, the lessons from this case are directly applicable across the accountancy profession.

• Use AI as a support tool—not a substitute. AI can assist with drafting and initial research, but it should never replace professional analysis or judgement.
• Verify all technical content. Any reliance on statutory provisions, case law or regulatory requirements must be checked against authoritative sources, such as legislation databases or official guidance.
• Maintain robust supervision. Those responsible for reviewing work must engage meaningfully with its content. The presence of AI doesn’t reduce the standard of review required.
• Be transparent about process. Firms should ensure that the use of AI is understood internally and that appropriate controls are in place, including clear policies and training.
• Respond to errors with candour. If an error is identified, it should be investigated promptly and addressed openly. Attempting to rationalise or minimise the issue risks further breaches of professional duty.

Conclusion

This case illustrates the intersection between emerging technology and established professional standards. It demonstrates that while AI offers significant opportunities, it also introduces new and sometimes subtle risks.

For accountancy professionals, the message is clear: professional judgement, ethical conduct and accountability remain paramount. AI may inform your work—but it cannot replace your responsibility for it.

Those who embrace AI must do so with care, ensuring that its use enhances, not undermines, the integrity of their practice.