Courts in the US and UK are increasingly confronting a specific failure mode of generative AI: hallucinations, in forms of confident sounding legal citations or quotations that simply do not exist.

Two recent, widely reported incidents show that judges are treating unverified AI output not as a quirky tech mistake, but as a governance and professional responsibility breakdown. 

Two cases that crystallise the risk 

In Withers v. City of Aberdeen, Senior Judge Sharion Aycock issued a sanctions order after filings on both sides contained 'hallucinatory citations'. AI was used, outputs were not verified, and local counsel effectively became a 'rubber stamp'. The result was operationally severe, with pro hac vice admissions revoked, disqualification of local counsel, fines, and with the scheduled trial cancelled. 

Media coverage emphasised how unusual it is to see both parties’ lawyers caught in the same matter, and how this is becoming a systemic workflow risk. 

In the UK, the Law Gazette reported that Pinsent Masons and three solicitors referred themselves to the Solicitors Regulation Authority (SRA) after the court was misled by AI-generated 'hallucinations' in the matter of Cork & Anor v Smith (an Insolvency and Companies Court proceeding). The judge’s criticism focused more on supervision, verification, and honesty to the court, including a second AI-assisted explanation that was still incorrect. 

A separate legal analysis summarised the case and framed it as a warning about a “cavalier attitude” to accuracy when AI is introduced into legal drafting. 

Why insurers should care 

1. Litigation operations risk becomes insurance risk

Insurers are litigation engines: coverage disputes, defence panel management, subrogation, employment matters. If outside counsel files AI hallucinated authority, the insurer can face delays, adverse procedural outcomes, higher defence costs, and settlement pressure. The Withers order is an example of how quickly the court can lose trust in the counsel’s work. 

2. Panel counsel oversight is now part of AI governance

The UK regulator’s materials on effective supervision reinforce that supervision is risk based and must be demonstrably effective, exactly the theme judges are highlighting when AI use is involved. For insurers, this maps neatly onto panel counsel guidelines, audit rights, and billing controls. 

3. Professional liability and E&O exposure will evolve

As lawyers embed generative AI into research and drafting, insurers writing lawyers professional liability, E&O, and even cyber will see more claims and more difficult causation arguments (human error vs. tool failure vs. supervision failure). That underwriting pressure is already visible in broader market moves to clarify or limit AI related liabilities in commercial policies.

Implications for insurers

These cases are best read as early signals of a new normal: courts and regulators are converging on the idea that AI output is never a substitute for professional judgment, and that failures will be treated as process failures (verification, supervision, controls), not 'technology surprises'. For insurers, that means AI risk management has to extend beyond underwriting models and into litigation operations, third party oversight, and policy language strategy.