Attorneys are being sanctioned for submitting legal citations that do not exist. And AI is at the center of it.
On the latest episode of The Byron Martin Show, Bret Uhrich, Partner at Walker Heye, walks through real cases where AI-generated information made its way into legal work without being properly verified. The consequences are a reminder that even when AI sounds confident, the person using it is still responsible for getting it right.
The conversation is a case study in why "AI is fantastic" and "AI needs guardrails" aren't contradictory statements. They're both true at once.
Bret pointed Byron to a database tracking court cases involving AI-generated legal errors, and the pattern is clear: this isn't a rare, one-off embarrassment anymore. It's a recurring event. As he put it, the first major case of a lawyer citing fake, AI-hallucinated cases hit legal press in early 2023, when courts were still figuring out how to respond. Now, similar cases are surfacing every few days.
One recent case stood out: four attorneys across multiple firms were sanctioned after hallucinated citations made their way into court filings. Some attorneys had used AI, while others were held responsible for failing to adequately verify the work before putting their names on it.
And it wasn't an isolated example. Bret described another matter where an expert declaration contained false AI-generated citations, and another where a supposed “three layers of review” process still failed to catch false citations.
The consequences can go far beyond a lawyer's reputation. Byron and Bret also discussed a case where plaintiffs and defendants were left without legal representation and given 60 days to find new counsel. That's no small disruption in litigation that can already take months or even years to resolve.
Perhaps the most striking part of the conversation was how courts are responding to attorneys who plead ignorance. In one case Bret referenced, an attorney claimed she didn't even understand what an AI "hallucination" was, even after the court had made her aware of the problem, and she continued using the same AI tool to produce legal documents. The court was not persuaded. It found that she knew, or reasonably should have known, about the risk and acted in bad faith by failing to verify the legal authority she submitted.
That helps illustrate how the legal system’s response is evolving. Bret noted that one of the first major AI-citation cases in early 2023 resulted in a sanction of about $5,000 and drew significant attention in the legal press. Since then, he said courts have become less and less accepting of the “I didn’t know how AI works” excuse. The lesson for professionals is clear: using AI does not remove your responsibility to understand its risks and verify the work it produces.
Bret shared the story behind Subnautica 2, the sequel to the popular underwater survival game. Krafton had acquired the game's developer, Unknown Worlds Entertainment, for $500 million, with the agreement including up to another $250 million in earnout payments if certain revenue targets were met.
As projections showed Subnautica 2 could trigger a significant earnout, Krafton's CEO sought ways to avoid the obligation. After receiving warnings about the legal and reputational risks, he turned to ChatGPT for another perspective. The chatbot generated a strategy for what to do if the company couldn't reach a deal. Krafton ultimately pursued a takeover strategy, terminated key leaders, and ended up in court. The Delaware Court of Chancery later found that Krafton breached the acquisition agreement.
The common denominator, Byron noted, is a false sense of confidence. People can begin treating AI output as a verified answer rather than a starting point, relying on the tool instead of stopping to think critically about what it’s telling them.
He compared it to "lemmings walking off a cliff". The tool says something confidently, so the decision-maker stops asking whether it's actually the smart move.
One of the more nuanced points in the discussion was about who is most likely to misuse AI, and it's not who you'd expect. Bret pointed out that attorneys who already enjoy legal research and writing, and feel confident in those skills may be less likely to lean heavily on AI. The people most attracted to these tools may be those who see AI as a way to make tasks they dislike or struggle with easier. But those may also be the situations where careful review matters most.
Bret also cited a Stanford study from 2024 that examined hallucination and error rates across general-purpose AI and legal-specific AI tools. While the legal-focused platforms performed better in some areas, the study still found meaningful rates of hallucinated and incomplete answers. The tools continue to improve, but the findings reinforce Bret’s larger point: AI-generated legal work still requires human review and critical thinking.
Bret was clear that AI's shortcomings aren't just about factual accuracy. Legal reasoning presents it's own challenges. In their conversation, Byron and Bret discussed examples of AI struggling to distinguish what a party argued in a case from what the court actually held, misunderstanding important legal principles, or failing to recognize what actually constitutes the holding of a case.
AI can process enormous amounts of information and produce convincing answers quickly. But as Byron and Bret discussed, it can't replace the critical thinking required to interpret that information and determine whether the answer is actually right. That's why human expertise and judgment still matter, especially when the consequences of getting it wrong are significant.
Beyond accuracy, Bret flagged a second, quieter risk: confidentiality. Under Washington's Rules of Professional Conduct, the definition of confidential client information is broad. Broad enough that even a client's name can qualify.
That creates another consideration for attorneys using AI. If client information is entered into an AI system that isn't closed-loop, it can potentially create a confidentiality issue under the rules governing the attorney-client relationship. And that risk exists regardless of whether the answer AI produces is right or wrong.
Whether you're running a law firm or any other business, Bret's conversation points to a few practical guardrails:
AI isn't going away, and the conversation isn't about avoiding it. It's about learning how to use it responsibly. Bret compared where we are with AI today to the early days of aviation. The technology will continue to improve, but getting there requires patience, safeguards, and a willingness to learn from what goes wrong along the way.
For businesses, the opportunity is significant. But so is the need for human judgment, clear guardrails, and critical thinking.
Teknologize is a SOC 2 Type I accredited Managed IT and Cybersecurity provider serving small to mid-sized businesses across Washington and Oregon. We deliver full-service Managed IT Support, Co-Managed IT Support, advanced Cybersecurity Solutions, and IT Compliance Services for regulated industries, including Healthcare, Financial Institutions, the Utilities Sector, Manufacturing, and Professional Services.
👉 Book a Discovery Call to see how Teknologize can support your business.
Our Offices
Tri-Cities, Washington – 509.396.6640 | Yakima, Washington – 509.396.6640
Bend, Oregon – 541.848.6072 | Seattle, Washington – 206.743.0981
Questions about your IT or Cybersecurity? Give us a call today!
Byron Martin: Aug 24, 2026
Attorneys are being sanctioned for submitting legal citations that do not exist. And AI is at the center of it.
Byron Martin: Jul 28, 2026
Every profession shapes us. The best leaders make sure it doesn't define them.
Byron Martin: Jul 16, 2026
For years, budgeting for business computers was fairly predictable. You could estimate replacement costs, plan refresh cycles, and expect only modest...
Byron Martin