Reuters: AI errors in court become a career risk for lawyers

In Alabama, a federal judge has suspended a lawyer from practicing in his court for six months. The issue was not only that a legal filing contained false quotations. The judge also found that the lawyer obstructed the inquiry into whether generative AI had been used to draft the filing.

Reuters reports that U.S. District Judge Harold Mooty III in Huntsville sanctioned attorney H. Gregory Harp after a brief contained four quotations that did not appear in the cases it cited. Harp also deleted his ChatGPT account just days after the court ordered him to produce records from it as part of the inquiry.

That is the part executives should pay attention to. The false citations matter. But the sharper governance problem is what happens when AI use becomes a matter of evidence, auditability and professional trust.

According to Reuters, the judge warned that any lawyer who believes the court will hesitate to investigate and impose “career-altering sanctions” is gravely mistaken. Harp was also disqualified from representing the plaintiff in the underlying case, an employment lawsuit against a bank.

This is not just a lawyer story

It is tempting to file this as another story about hallucinated AI in court. That is too narrow.

The operational lesson is that generative AI is moving from writing assistance into auditable work surfaces. When an employee, adviser or external law firm uses ChatGPT, Claude, Copilot or a sector-specific AI tool for work that may later end up in court, regulatory review, audit or an internal investigation, the prompts, outputs, source checks and account deletion practices become part of the risk picture.

For companies, this is highly relevant to legal work, contracts, HR disputes, public procurement, compliance, board papers and regulatory reporting. It is not enough to say that AI may be used as long as the result is checked. The real questions are who checks it, how the check is documented, and what logs will still exist if the matter is challenged later.

Reuters notes a broader trend in U.S. courts: judges are increasingly considering how to sanction lawyers when filings contain AI-like errors such as fabricated case citations or misquoted material, including when the lawyers deny that AI was involved. AI use is not prohibited. Lawyers are still bound to verify the accuracy of what they submit.

That distinction travels well beyond the United States. The technology may be allowed. Uncontrolled use is not automatically defensible.

What boards and management should demand

The first requirement is to separate casual writing support from work with evidentiary or regulatory value. An executive using AI to tighten the language of an internal note is one thing. Using the same tool to draft legal analysis, dismissal documents, tender material, board submissions or filings to authorities is another.

The second requirement is logging and retention. If the organization allows AI in legal or regulatory workflows, it must know whether the tool stores history, who can access it, where the data sits, when content can be deleted and how it can be exported in a dispute or investigation. That needs to be settled before the file is on fire.

The third requirement is source verification. AI-generated references must not merely look plausible. They must be checked against primary sources. In legal work, that means court rulings, statutes, preparatory works, contract text or another authoritative source. In executive work, it means that figures, commitments, regulatory requirements and external claims must be traceable.

The fourth requirement is clear accountability. It cannot be unclear whether responsibility sits with the junior employee who prompted the tool, the partner who approved the work, the vendor that supplied the system or the manager who demanded speed. AI can accelerate work. It cannot own the signature.

Vendor governance gets more concrete

The case also changes the conversation around procurement of legal and advisory services. Many companies will increasingly receive work where AI has been used in the background, even when that is not visible in the headline.

Contracts should therefore become more precise. Ask providers which AI tools may be used, what client data may be entered, whether data is used for model training, how outputs are quality-controlled, which logs are retained and which subcontractors have access. Require unauthorized or non-compliant AI use to be reported as an incident, not buried as a footnote.

Internal policies must also be practical enough to be followed. A blanket ban often creates shadow use. Free use without controls creates evidentiary chaos. The better route is approved tools, clear zones for what AI may and may not be used for, mandatory source verification, and a low threshold for escalation when AI has been involved in work that could have legal consequences.

The executive takeaway

Most companies are not bound by an Alabama federal judge. But the reaction shows how quickly AI use can shift from productivity gain to credibility problem.

The same pattern can hit corporate teams. A leadership team that approves AI in legal work without traceability is taking a risk that only becomes visible when something goes wrong. By then it may be too late to discover that prompts were deleted, the provider cannot document verification, or nobody knows who actually checked the content.

This is not mainly a warning against ChatGPT. It is a warning against weak governance of AI in work that depends on trust.

Sources and media

• Primary source: Reuters, Mike Scarcella, May 22, 2026: https://www.reuters.com/legal/litigation/judge-says-lawyers-ai-use-risks-career-altering-consequences-2026-05-22/
• Court document: U.S. District Court order linked by Reuters: https://fingfx.thomsonreuters.com/gfx/legaldocs/myvmylewkvr/Alabama%20AI%20order.pdf
• Reuters image: Dado Ruvic/Illustration/File Photo via Reuters Connect. Not rehosted by hogby.ai.
• Thumbnail: OpenAI Image 2 / hogby.ai.