Confidently wrong: What generative AI means for your reputation, your obligations

Generative AI was used in the preparation of the following commentary. All citations were verified and all analysis reflects the author’s own judgment. See, it is not hard to do.

A good legal reputation is not built from a single case or a handful of court appearances. It is amassed over years through competence that clients can trust, diligence that ensures hundreds of deadlines are met and thousands of phone calls are returned, and candor that courts can rely on in every statement of fact and law.

It is earned through the consistent exercise of those qualities across every matter and every interaction. A reputation is the sum of countless moments accumulated over the arc of a career. The careless use of generative artificial intelligence can destroy it overnight.

Generative AI is the most consequential tool the legal profession has ever encountered. Unlike prior legal technology, it both gathers information and produces finished work product. That is both the source of its promise and its danger.

Prior technology, like Lexis and Westlaw before it, delivered raw material and left the lawyering to the lawyer. Generative AI collapses that division. Nothing in the final product shows which parts can be trusted and which cannot. It produces output that is polished, authoritative-sounding, and, at times, wrong in ways that are not evident on its face.

The lawyering that the tool cannot supply for itself is exactly what the professional conduct rules have always required, and this technology now demands it in ways the profession has not previously encountered.

Rather than attempting to construct a new regulatory framework around a technology that continues to evolve at a pace no rulemaking process could reliably track, the Supreme Judicial Court grounded its generative AI guidance in the existing Massachusetts Rules of Professional Conduct.

The professional conduct rules are durable precisely because they are principle-based rather than technology-specific. They set a floor, not a ceiling, for what the profession expects of every lawyer practicing in the commonwealth. However AI is to be used in legal practice in the years ahead, the obligations of competence, diligence and candor built into that floor will not become obsolete.

That durability is the point. Competent representation, under Mass. R. Prof. C. 1.1, requires “the legal knowledge, skill, thoroughness, and preparation reasonably necessary for the representation.”

Comment 8 to that rule makes the technology dimension of that obligation explicit: a lawyer should “keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology, and engage in continuing study and education.”

That comment was written before generative AI existed, but it reads as though it was written for this exact moment. Notice that it requires weighing benefits and risks, not avoiding one in favor of the other. A lawyer who uses generative AI without understanding it has not satisfied that obligation. Alternatively, a lawyer’s decision not to use generative AI should rest on an informed understanding of its capabilities and limitations rather than unfamiliarity alone.

Diligence requirements carry the obligation further. Pursuant to Mass. R. Prof. C. 1.3, a lawyer must “act with reasonable diligence and promptness in representing a client.” Generative AI can deliver the promptness part of that obligation in seconds. It does nothing for the diligence half, which remains entirely the lawyer’s to perform. The tool can produce the work product, but only the lawyer has a reputation to lose if it is wrong.

Candor imposes an additional obligation. Mass. R. Prof. C. 3.3 prohibits attorneys from making false statements of fact or law to a tribunal and requires correction when such statements occur. Use of this emerging technology does not alter that duty; it heightens the need for careful verification before any AI-assisted work product is submitted to a court.

The stakes have been made clear by disciplinary proceedings, sanctions orders, and denials of pro hac vice admission. Attorneys risk fines, public reprimand, potential suspension, and damaged reputations.

When output and outcomes are flawed, and no one in the chain of preparation understood the technology well enough to catch it, the damage is not only to the attorney’s reputation. It is to the reputation of the bar at large.

The offending conduct has come in two forms. The first is the hallucinated citation, a reference to a case that does not exist. The second, more dangerous, is the fabricated proposition: a citation to an actual case, the holding of which is wholly an AI invention dressed as precedent. Both are confidently wrong and neither announces it.

The common thread is not bad faith. It is an attorney who used the tool without understanding it, or who signed a filing without verifying what it produced. Generative AI does not change what competence and diligence require. It only raises the cost of failing to meet them.

The requirement is straightforward. An attorney using generative AI in legal work has to understand how the tool produces its output and where that output can go wrong. An attorney who chooses not to use it has to be making that choice on the same basis — an informed judgment about the technology’s capabilities and limitations, not unfamiliarity or discomfort.

The legal technology market is already responding to part of this problem. Retrieval-augmented generation, which grounds AI output in verified legal databases rather than generating from training data alone, meaningfully reduces the risk of a hallucinated citation. It does nothing to prevent a fabricated holding attached to a real case. No tool, however capable, absolves a lawyer of that professional judgment.

Law schools across the country are recognizing this urgency. The majority of law schools nationally now offer dedicated AI coursework, and a growing number require it of every student. The technological landscape will keep moving. The Rules of Professional Conduct, steady and unmoved by any of it, will keep requiring the same thing they always have. The schools are adapting their curricula to that landscape in a way that reflects those enduring requirements.

Attorneys already in the profession have the same obligation, and they risk their reputation by failing to meet it.

One practice worth consideration by practitioners who use AI tools in their work is noting that fact in their filings. Doing so is not a requirement under current SJC guidance, but it is consistent with the candor obligations the Rules of Professional Conduct demand.

A filing prepared with the assistance of generative artificial intelligence should include a certification that generative AI was used and that all factual assertions, legal arguments, and citations have been personally reviewed and verified for accuracy by the filer. This exceeds what the rules currently require and invites the kind of scrutiny responsible AI use should be able to withstand.

The Rules of Professional Conduct set the floor. What protects a reputation above that floor is the judgment to understand the tools being used and the discipline to verify what they produce before those results are submitted to a court or a client.

The stakes, however, extend beyond any individual attorney’s standing. The credibility of the legal system depends on the accuracy of what is produced. When output and outcomes are flawed, and no one in the chain of preparation understood the technology well enough to catch it, the damage is not only to the attorney’s reputation. It is to the reputation of the bar at large, the assurance that it can be trusted with matters that carry serious consequences for the people who depend on quality legal counsel. That is a reputation worth protecting.

Judge Brian Palmucci sits on the District Court bench. The views expressed above are his own and not offered in his capacity as a sitting judge, nor do they represent the position of the District Court or the Trial Court.