If AI writes the first draft, where does the next generation of lawyers learn to think through the case?
U.S. District Judge Arun Subramanian declined to sanction lawyers over an erroneous AI-assisted filing, Reuters reported on October 2. He nevertheless urged verification and questioned how firms will train junior lawyers as AI use grows, suggesting that initial drafts could be prepared without AI. Harvey tools assisted the filing; there is no indication the client required their use.
The episode raises two distinct concerns for legal practice: whether a document is accurate enough to file, and whether the process of producing it develops the lawyer’s judgment.
The firm had an AI policy and mandatory training
In a declaration filed in Hill v. Foundation Media LLC, Shapiro Arato Bach partner Cynthia Arato acknowledged that unchecked AI use produced incorrect quotations and a citation to a nonexistent case.
Arato said the firm had conducted training, including a mandatory session with outside ethics counsel, and required employees to sign an acceptable-use policy. That policy required AI output to be reviewed, validated and corrected before use. Her declaration described the conduct as contrary to the firm’s policies and training.
That account makes the episode particularly relevant to firms that have already introduced formal AI safeguards. A written policy establishes expectations. The filing process still needs to make those expectations effective.
Verification and training solve different problems
Checking citations addresses an immediate risk. A reviewer can locate the judgment, confirm the quotation and assess whether the authority supports the proposition advanced.
Training asks a broader question. Can the junior lawyer identify the strongest argument, recognize an adverse authority and explain why the proposed structure serves the client’s case?
A document may survive citation checks while its author has done little of that reasoning. Conversely, a lawyer who works through a difficult first draft may gain valuable experience even when the draft needs extensive revision.
The management challenge is deciding which parts of the assignment should provide that experience.
Why Canadian law firms should pay attention
The Law Society of Ontario’s April 2024 guidance connects generative AI use with existing obligations of competence and supervision. It emphasizes that lawyers remain professionally responsible for their work and must exercise independent judgment when using technological assistance. It also recommends clear workplace policies about when AI may be used in delegated assignments.
Subramanian’s comments add a useful training question to that framework: what should an associate be expected to understand before relying on a generated draft?
One possible approach is to require the associate to prepare an issue outline and identify the principal authorities before using AI. Another is to reserve selected first drafts for independent work, followed by a discussion with the supervising lawyer about structure, omissions and strategic choices.
These are practice options, rather than requirements imposed on Canadian firms by this U.S. proceeding.
What does the supervising lawyer need to see?
A polished draft alone may reveal little about how it was produced. A short discussion can reveal much more: which authority is strongest, what the other side will argue, and where the analysis remains uncertain.
For Canadian firms, that offers a concrete way to connect AI adoption with professional development. Review the work product, but also ask the lawyer to explain the reasoning behind it.
The useful discussion for a practice group is therefore specific: which drafting tasks should junior lawyers complete themselves to develop the judgment they will later need to supervise AI-assisted work?
Sources: Reuters’ October 2 report , the court’s order requiring counsel’s AI declaration and the Law Society of Ontario’s April 2024 white paper .
Related coverage: B.C. Court Warns That Poor AI Filings Can Undermine Access to Justice.