Let’s be direct about something the legal AI discourse is dancing around: using AI to generate work product you do not review, or are not qualified to review, isn’t a workflow shortcut. It’s malpractice.
That’s a strong word, and it’s the right one. The junior associate analogy for AI has a second failure point that makes this clear, and it’s harder to talk about than the technology itself because it implicates the people using it.
The analogy works, to the extent it does, because it assumes the supervision layer is intact. With a junior associate, judgment flows top-down. The supervising partner reads the work, catches the errors the junior attorney didn’t know to look for, and the system functions as designed. That review isn’t just a formality. It’s the entire mechanism by which the client is protected and the profession maintains its standard of care.
With AI, that layer fails in two distinct ways, and the profession is being honest about neither of them.
The first failure is diligence. Attorneys who generate work product with AI and submit it without meaningful review. Not skimming, not spot-checking, but actually reading it with the same critical eye they’d apply to a junior associate’s draft. This happens. It’s happening at scale. The speed and fluency of AI output creates a psychological pull toward trust that the technology hasn’t earned. The output looks right. The citations look real. The structure looks sound. So it goes out the door.
The second failure is competence. Attorneys, and increasingly non-attorneys operating in legal-adjacent roles, using AI in practice areas where they lack the underlying expertise to evaluate the output at all. This is more insidious because it’s invisible. The person reviewing the work doesn’t know what they don’t know. The errors aren’t caught because the reviewer isn’t positioned to catch them. The supervision layer exists on paper and nowhere else.
Both failures have the same name under Rule 1.1 of the Model Rules of Professional Conduct. Competence requires the legal knowledge, skill, thoroughness, and preparation reasonably necessary for the representation. The rule has been around since 1983. The technology is new. The standard isn’t.
The parallel to how the profession handles other supervision failures is instructive. A supervising partner who rubber-stamps a junior associate’s work without actually reading it doesn’t get to blame the associate when something goes wrong. The supervision obligation is theirs. It doesn’t transfer because they delegated. The same logic applies here, and the bar has been clear, if not always loud, on this point: using AI doesn’t relieve the attorney of responsibility for the work product it generates.
“The AI got it wrong” is not a defense. It’s the indictment.
The courts are making this concrete at a rate the profession can no longer ignore. Last year saw a rapid increase in sanctions against attorneys for filing briefs containing fictitious AI-generated citations, with one researcher tracking more than ten cases from ten different courts in a single day. Sullivan & Cromwell, one of the most prominent Wall Street firms in the country, apologized to a federal bankruptcy judge last week for submitting a filing containing AI hallucinations, including fabricated citations and misquoted law. Courts imposed more than $145,000 in sanctions in the first quarter of 2026 alone.
These aren’t cautionary tales about rogue solos cutting corners. Sullivan & Cromwell has more than 900 lawyers and a reputation built over generations. If it can happen there, the assumption that institutional size or reputational stakes provide a natural check is wrong.
What makes this conversation uncomfortable is that it requires the profession to reckon with something beyond the technology itself. The question isn’t just whether AI is reliable enough to use. It’s whether the people using it are in a position to evaluate the output, and whether they’re actually doing so. In too many cases, the answer to one or both of those questions is no. The honest response to that isn’t a better AI product. It’s a more rigorous application of the competence standard we already have.
The profession has largely moved past debating whether to adopt AI. The conversation now is around responsible implementation, and the courts are supplying the answer in the clearest possible terms. Competence has always been the rule. The technology doesn’t change that. It just makes the consequences of ignoring it harder to avoid.
Derek Francisco is a licensed attorney and fractional in-house counsel. He writes on law, legal technology, and the economics of institutional change. Views expressed are solely his own.