Judges 'debated' mandatory AI disclosure

Ms Justice Eileen Roberts has said that the emerging picture on AI use in common-law systems across the world is one of "cautious acceptance", rather than prohibition.  

She was speaking at the annual conference of the Law Society Litigation Committee in Blackhall Place (15 October). 

The High Court judge addressed the event on the background to the recent practice direction HC 142 on AI, which came into effect on 1 September.  

She told the conference that the twin aims of safeguarding judicial independence and maintaining public confidence in the administration of justice were at the heart of the Irish judiciary's approach to AI.

Court of Appeal guidelines 

Ms Justice Roberts referred to some cases in which AI had clearly been used inappropriately, the most high-profile of which was a Court of Appeal decision earlier this year in Guerin v O'Doherty. 

O'Doherty was a lay litigant in the case and used AI to draft her submissions, which led Ms Justice Costello, in her judgment, to set down some guiding principles for the use of AI in court proceedings.  

The Court of Appeal said that a litigant using AI to help with research was responsible for whatever was produced and could not rely on it without independently checking it. 

Ms Justice Roberts said that the judgment's commentary on disclosure of AI usage and potential sanctions had been "supplemented and clarified" in HC 142 and in a similar Court of Appeal practice direction CA 16. 

Acknowledging that there may be queries about aspects of the practice direction, she told the event that the judiciary was "learning as we go" and that it would be tweaked as necessary as time went on. 

Dividing line on disclosure 

Speaking about AI developments in justice systems globally, she said: "The real dividing line seems to be not whether AI can be used, but rather how much disclosure of its use is required, and whether evidentiary material should be treated differently from other legal submissions or pleadings." 

In England and Wales, the judge said, the focus was mainly on accountability and accuracy, rather than on mandatory disclosure, while Canada had taken a more structured approach, requiring parties to disclose when litigation documents contained content generated by AI. 

She saw Australia as the most regulated jurisdiction, citing the Supreme Court of New South Wales's ban on the use of AI to generate the content of affidavits, witness statements, or character references. Experts in that state also require leave of that court before AI can be deployed in an expert report. 

The High Court judge told the event that the judges did "consider carefully" mandatory disclosure of AI use, but came to the view that its use was now so ubiquitous that it would be "nonsensical" to have everybody declare its use in every case. 

'Responsible' AI use 

Ms Justice Roberts said that the Irish courts accepted that AI could improve efficiency and access to justice, but added that the practice direction had placed particular emphasis on evidentiary material. 

"We're slightly closer to the Australian model in relation to affidavits and witness statements, but we remain broadly aligned with England and Wales in permitting responsible use of AI for legal research and drafting," she told the conference. 

Use for research and drafting was subject to independent clarification and personal accountability, the judge said, but with no general requirement for mandatory disclosure. 

The potential sanctions for AI-related breaches set out in HC 142 include costs, applications being dismissed, evidence being discounted, and the possibility of disciplinary referrals. 

Referring to the O'Doherty case, Ms Justice Roberts said that the court had accepted that the defendant had not intended to mislead the court, but pointed out that the judgment had been delivered before there were published guidelines or the practice direction had come into effect. 

"I would say, with virtual certainty, that anybody else that comes in now with the same set of circumstances is very unlikely to get the same level of leniency," she stated. 

Concerns on witness statements 

Under the practice direction, affidavits and witness statements, which are intended to stand as evidence, must include a declaration that generative AI has not been used to generate or alter their substantive content. 

Experts' reports must contain a similar declaration or, if they do use generative AI in a material way, they are required to disclose exactly how they have used it. 

The judge said that the reason for the focus on evidential material was linked to pre-AI concerns about how witness statements and expert reports were generated – such as who drafted them and what instructions were given. 

"In most cases, the law is actually pretty clear, but what makes the difference to the outcome of the case is the evidence; hence, we've got to be sure that the evidence that we are dealing with is accurate, is real, and is authentic," she told the conference. 

Too often, she added, there was "little correlation" between a witness statement and the person who gave oral evidence on what was in their witness statement. 

The judge said that AI had the potential to make that situation "even more difficult".  

She told the conference that the judiciary did not intend the practice direction to disallow the use of AI to assist with the organisation of source material, such as formatting or transcription of materials from verified source documents.

Lessons on AI adoption 

The AI section of the conference also heard from RDJ partner Ricky Kelly, who told the event about his firm's experience of adopting the technology and the lessons it had learned from the process. 

Kelly said that the firm had begun looking at AI systems because of competitive pressures, client expectations, and a realisation that the technology was not going away. 

Some of the key questions for RDJ included how the provider addressed concerns about the profession's confidentiality and privilege obligations, as well as what steps could be taken to minimise 'hallucinations'. 

‘Invest in proper training’ 

Kelly said that some of the key lessons for his firm included: 

• Draft a firm-wide AI policy, 
• Use a formal framework when evaluating AI tools, 
• Redesign research workflow to include mandatory verification, and 
• Invest in proper training. 

He also warned that the technology was "not cheap" and that the cost model for AI technology was likely to change. 

Using good judgement 

Paul Breslin (PPC course manager, Law Society) told the event that AI was interwoven through the Law School's curriculum. 

This included an 'AI in legal practice' module that started in 2025, covering usage, risks, limitations, and the regulatory and ethical frameworks for the technology. 

Breslin stressed the importance of developing trainees' judgement on the appropriate use of AI.