What does it mean to be AI literate as a lawyer in 2026?

That foundational question kicked off the Thomson Reuters Institute's AI and Future of Legal Practice webinar series in September in the webinar, AI 101 & Why AI Creates Ethical Dilemmas for Lawyers.

Webinar moderator Megan Carpenter, former dean of the University of New Hampshire School of Law, framed the conversation around AI literacy for lawyers as one of three interlocking themes: technical literacy and errors; ethical dilemmas; and the evolving culture of the legal profession, including education.

Indeed, Carpenter said, training is "not just for academics anymore,”  but rather than "learning and skill-building around AI is for all lawyers and faculty." This framework set the tone for a discussion with panelists that moved from the mechanics of large language models to the real consequences of getting AI wrong in practice.

Understanding the black box of AI and how it works

John Hudzina, lead research scientist at Thomson Reuters Labs, began the webinar with a brief look at the origins of the advanced technology that is impacting us now, noting that today's frontier AI models are not as conceptually different from Eliza, which was a rudimentary chatbot from the mid-1960s, as people might assume.

The key difference, Hudzina pointed out, is that modern systems use statistical prediction rather than simple keyword rules to generate text one token at a time based on probability, and sophisticated multi-agent systems can appear to "check their own work." However, they are still just making predictions, he added, and there is no guarantee of accuracy, no matter how many verification loops are built in.

And this technical foundation matters because it reframes what AI literacy should mean for practicing lawyers, said webinar participant Michael Yang, a principal at Husch Blackwell Consulting. Any understanding of AI should take into consideration ABA Model Rule 1.1's duty of competence, which argues that AI literacy is not a technical requirement, but rather one that is grounded in knowledge and preparation.

Indeed, Yang explained, lawyers do not need to be engineers, but they do need to understand what a tool does and does not do — in addition to understanding how it can go right and how it can go wrong. Trust in AI output should scale with risk, he noted. For example, using a tool to polish an email is low stakes, but using it to generate citations for a court filing carries far higher consequences if the tool gets it wrong.

Check out our next webinar on Oct. 15 on Effective Verification Habits for law students and lawyers

Human review is more than proofreading

Perhaps the most striking insight of the webinar was Yang's explanation of why AI errors are so hard to catch. Unlike a human who might hedge or signal uncertainty, AI systems present nearly everything with the same assured tone, meaning "the moment it is wrong, it's going to be confidently wrong because of how they approach it." This creates a dangerous illusion of reliability, Yang said, especially when lawyers conflate "human review" with simple proofreading.

Yang was emphatic that genuine review requires subject-matter understanding rather than just a glance. For instance, a human can stare at a document written in a language they don't understand and never catch what is wrong with it, and the same is true when they are reviewing AI output in unfamiliar territory.

This is compounded by scale. Hallucinations — or false “facts” generated by generative AI systems and can occur due a number of issues — are not one uniform error. Indeed, they range from misapplied jurisdictions to misconstrued holdings; and when a tool generates 50 pages of confident, plausible-sounding text, spotting the flaw buried inside of it becomes exponentially harder than catching a single wrong answer.

Real-world consequences are already mounting as databases tracking AI-related hallucination incidents in court filings show the problem growing even as awareness increases. In addition, courts have been unambiguous that this responsibility rests with the lawyer, not the tool, regardless of who, or what, drafted the language.

Rethinking legal education and the continuous-learning profession

Webinar participant Nicole Phillips, Associate Professor of Legal Research, Writing, and Analysis at the University of San Francisco School of Law, brought the conversation into the classroom. Phillips said that for her, AI literacy in legal education is about developing judgment to understand the tools well enough "to be able to make thoughtful decisions about when and how to use them, and when and how not to use them."

Rather than banning or blanket-permitting AI, Phillips said she builds assignment-specific policies that map to the exact skill she is teaching at each stage of a course. She then requires students to be able to justify their tool choices to mirror the same explainability standard Yang described for practicing attorneys.

Q: What does overreliance on AI mean for lawyers?

A: Overreliance on AI occurs when the tool stops helping someone think and starts doing the thinking for them, a standard that applies equally to law students and seasoned practitioners.

Phillips offered a clarifying definition of overreliance that applies equally to students and seasoned lawyers. “When the tool stops helping someone think and starts doing the thinking for them,” that is the distinction between AI augmenting thinking and judgment and replacing it. This perspective emerged as the throughline connecting all four speakers' perspectives on the topic of overreliance.

Looking ahead

Whether in a classroom or a courtroom, the webinar participants agreed that AI literacy, ethical vigilance, and new educational approaches are facets of the same AI transformation that is currently reshaping the legal practice.

As such, today's lawyers need to understand how these AI tools generate predictions, where they tend to fail, and when human judgment must step in as the new baseline for competent practice in the AI era.

The Thomson Reuters Institute's AI & the Future of Legal Practice series continues this conversation on October 15 with a session on effective verification habits, followed by a November 19 session on what effective human-in-the-loop practice looks like.