AI may reduce the time required to produce a brief or opinion. The relevant measure, however, is not the cost of the first draft. It is the cost and reliability of reaching a sound result while developing lawyers who can reach the next one.
The most important observation in a recent federal order about artificial intelligence was not about a fabricated case—it was about the lawyers who are expected to catch one. In Hill v. Foundation Media LLC, U.S. District Judge Arun Subramanian warned that overreliance on AI may stunt the development of young lawyers and ultimately harm clients. He suggested that firms consider having lawyers prepare an initial draft without AI assistance.
The order identifies a problem that ordinary cite-checking instructions do not solve. A lawyer cannot reliably test an AI-generated argument without enough experience to recognize what is missing, overstated or legally implausible. If AI displaces too much of the research and drafting through which that judgment develops, the supposed human safeguard becomes less dependable.
For legal departments, this is not simply a law firm training concern. It affects who will understand the record, identify adverse authority and exercise judgment when the machine produces a polished but wrong answer.
What Happened in Hill
Hill v. Foundation Media LLC, No. 25-CV-5947 (AS) (S.D.N.Y.), was a copyright action brought by a self-represented plaintiff. On September 22, 2026, Judge Subramanian adopted a report and recommendation dismissing the complaint with prejudice. No. 25-CV-5947 (AS) (S.D.N.Y. Sept. 22, 2026). Although the defendant prevailed, the court separately required lead counsel to declare whether AI had been used to prepare a response brief and whether identified quotations and a case citation resulted from unchecked AI use. No. 25-CV-5947 (AS) (S.D.N.Y. Sept. 24, 2026).
After counsel responded, the court declined to impose sanctions and expressed its hope that the lawyer principally responsible would not be punished. The September 29 order nevertheless called the situation a “wakeup call.” It directed lead counsel using AI to confirm that every citation, quotation and legal proposition has been double-checked, noting that even advanced large language models remain capable of difficult-to-detect hallucinations. No. 25-CV-5947 (AS) (S.D.N.Y. Sept. 29, 2026).
Judge Subramanian then turned to clients and training. He cautioned clients against requiring AI-assisted brief writing merely to reduce expenses and warned that overreliance on AI may stunt the development of young lawyers. Judge Subramanian emphasized that practitioners should consider ways to ensure necessary training of young lawyers as this technology develops, such as human preparation of an initial draft. Id.
Why the Training Concern Is Not Abstract
Two appellate disciplinary decisions illustrate what can happen when drafting is separated from informed review. They do not treat AI use as misconduct. They focus on the lawyer who signs and files work that they have not adequately checked.
In Lnu v. Blanche, the Ninth Circuit sanctioned two lawyers whose briefs included nonexistent cases, misattributed quotations and distorted accounts of real cases. 177 F.4th 1014 (9th Cir. 2026). Unlicensed law graduates had prepared legal arguments and gathered authorities, and the supervising lawyers acknowledged that they did not ordinarily verify the citations. Each lawyer received a $2,500 personal sanction and a six-month suspension, together with notice obligations and two years of AI-use certification.
Fletcher v. Experian Info. Sols. Inc. involved a similar scenario. 168 F.4th 231 (5th Cir. 2026). Counsel used AI to prepare a substantial portion, if not all, of a reply brief, failed to verify the result and was not forthcoming when the court asked what had occurred. The court imposed a $2,500 sanction.
Lnu is especially instructive. The Ninth Circuit said the violation occurred when the lawyers signed and filed the briefs, not when someone used AI for research or drafting. The professional failure was submitting work without the understanding and verification necessary to own it.
Verification Depends on Experience
Assignments for young lawyers produce work for the client, but they also teach the lawyer how to recognize a legally significant fact, test an analogy, confront contrary authority and turn research into advice. Generative AI can accelerate those assignments. It can also remove the critical part that requires the lawyer to struggle with the authorities and decide what they mean.
That produces a circular risk: The workflow relies on a human reviewer to catch a convincing error while removing the work through which the reviewer would learn to recognize it. Telling a lawyer to verify the answer is inadequate if they do not yet know what a sound answer should look like.
This does not mean that every first draft belongs in an AI-free zone. The sensible dividing line is the purpose and risk of the assignment. Work that teaches issue identification, authority selection or case theory deserves more human effort at the front end than routine work whose accuracy can be checked mechanically.
The Existing Rules Are Sufficient
American Bar Association Formal Opinion 512 is advisory, but it gives firms a useful map. Competence still requires legal knowledge and judgment. Model Rules of Professional Conduct 5.1 and 5.3 still require reasonable supervision of lawyers and nonlawyers. Candor, confidentiality, communication and reasonable-fee duties still apply to the finished legal service. None of those obligations move to the generative AI software provider.
The harder question is operational: How should a firm receive the benefit of AI’s efficiencies without depriving its lawyers of the experience needed to supervise the work? Hill does not answer that question, but it makes the question difficult to ignore.
Questions Legal Departments Should Ask
Who First Analyzed the Problem?
For a dispositive motion, expert issue or significant opinion, determine whether a lawyer framed the issue and selected the authorities before AI began producing content. Human-first analysis need not mean human-only drafting.
Who Can Explain and Defend the Work?
The responsible lawyer should be able to explain what the cited authorities control, what contrary law was considered and how the record supports each material assertion. A signature should reflect comprehension, not simply approval of an edited output.
What Work Is Reserved for Development?
Ask how junior lawyers receive assignments that require them to find, read and synthesize primary authority before seeing an AI answer. The point is not to preserve inefficient habits; it is to build the judgment that later makes efficient review possible.
What Happens When the Tool Is Wrong?
A workable protocol should identify the affected work, preserve relevant prompts and outputs, assign responsibility for correction and address notice to the client or tribunal when required. Speed of correction and candor matter as much as prevention.
The Limits of the Decisions
Hill is a district-court order, and its discussion of lawyer development is advisory. It imposed no sanction and created no general rule that initial drafts must be written without AI. Lnu and Fletcher concern inaccurate filings and failures of verification or candor; neither treats junior lawyer development as an independent ground for discipline.
The decisions nevertheless expose a weakness in a common assurance about AI-assisted legal work: that a lawyer will always check the result. That assurance is only as good as the lawyer’s ability to recognize a defective answer.
A Better Measure of Efficiency
AI may reduce the time required to produce a brief or opinion. The relevant measure, however, is not the cost of the first draft. It is the cost and reliability of reaching a sound result while developing lawyers who can reach the next one. A workflow that saves hours today but weakens the firm’s ability to detect error tomorrow has overstated its efficiency.
For More Information
If you have any questions about this Alert, please contact Samuel W. Apicelli, Chloe Vizzone, any of the attorneys in our Intellectual Property Practice Group or the attorney in the firm with whom you are regularly in contact.
Disclaimer: This Alert has been prepared and published for informational purposes only and is not offered, nor should be construed, as legal advice. For more information, please see the firm's full disclaimer.