When AI Gets it Wrong: Lessons from Guerin v O’Doherty

Case Background

On 26 March 2026, the Court of Appeal dismissed an appeal brought by journalist Gemma O’Doherty, affirming an earlier High Court decision refusing to strike out defamation proceedings initiated by James Guerin, brother of the late journalist, Veronica Guerin. The proceedings arise from a series of social media posts and online video broadcasts published in 2019.

In dismissing the appeal, the Court of Appeal addressed, among other things, the use of GenAI in the preparation of Ms O’Doherty’s written submissions. The defendant, acting as a litigant in person, relied on AI-generated material which included references to authorities that did not exist.

The Court described these as “hallucinations” and noted that their inclusion required the opposing party to spend time and resources attempting to locate these “fake” cases, resulting in unnecessary cost and inefficiency.

Judicial Guidance from Costello J

Read alongside the Law Society of Ireland’s Guidelines for the Use of Generative Artificial Intelligence (December 2025), the judgment of Costello J helps to crystallise what Irish litigators should now treat as minimum standards when using GenAI in contentious work, particularly around the risk of “hallucinated” authorities and the need for disclosure in the written submissions. The key principles identified by Costello J are as follows: 

• GenAI use is permissible, but not without responsibility.
• An obligation to verify all materials submitted to the Court.
• “Hallucinated” authorities are unacceptable.
• Duty of candour and transparency: including the need to disclose when GenAI tools have been used to the Court and opposing parties.
• Responsibility applies to all litigants (i.e. legally represented parties and litigants in person).
• Potential for adverse consequences where GenAI is being misused.
• Professional standards remain unchanged, including duties of honesty, competence and independence.

The Court’s Approach to GenAI in Submissions

The significant aspect of this judgment for practitioners is the Court of Appeal’s articulation of a short set of “principles of general application” governing the use of AI tools in litigation, as set out above. These do not prohibit the use of GenAI; rather they establish the conditions for its responsible use.

The Court confirmed that GenAI may be used to assist with legal research, provided it is used carefully and does not mislead the Court, whether deliberately or inadvertently, as in the Guerin v O’Doherty case. Where GenAI is used in the preparation of written submissions, there is a clear expectation that this should be disclosed to both the Court and opposing parties. Crucially, responsibility remains with the user: no authority should be cited unless it has been independently verified as existing and relevant.

This approach aligns closely with the Law Society’s guidance, which emphasises that generative AI must never compromise accuracy or professional standards. While those guidelines do not impose a general obligation to disclose the use of GenAI use in all circumstances, they reinforce the overriding duty not to mislead the Court and the need for independent review of authorities being used when preparing written sources.

Lessons for Practitioners

For contentious practitioners, the message is straightforward: while GenAI can be a useful tool, it does not dilute the duty not to mislead the Court and cannot substitute for professional judgement. The risks are particularly acute where AI-generated material purports to cite legal authorities, underscoring the importance of independent verification.

Practitioners should also be alive to privilege and disclosure risk when AI tools are used in the conduct of litigation. In particular, inputs to, and outputs from, open-access AI platforms may become difficult to protect and could materially broaden the scope of discoverable material. Relatedly, there is an emerging view that prompts used and results generated may be sought on discovery, potentially adding a further layer of cost and complexity to e-discovery exercises.

The judgment of Costello J in Guerin v O’Doherty does not prohibit the use of GenAI in Irish litigation. Rather, it provides a clear warning that its use must be accompanied by careful verification and appropriate transparency where materials are placed before the Court. Professional responsibility remains unchanged: while the tools may evolve, the duties owed to the Court do not.