CBA Record January-February 2026

THE YOUNG LAWYERS SECTION

nothing about AI suspends a lawyer’s duty to understand the tools they use and to supervise their outputs. The ABA made that explicit in Formal Opinion 512: Use generative tools if you like, but honor competence, confidentiality, communication, candor, supervision, and reasonable fees. It’s guidance, not gospel, and that’s the point. Courts, for their part, are converging on “use it, but own it.” Sanctions in Mata v. Avianca , 678 F. Supp. 3d 443 (S.D.N.Y. 2023), weren’t imposed because of what a robot wrote; they were imposed because humans failed to audit their bots. The humans let hallucinated citations into their briefs and then doubled down

human in the loop where originality matters. That’s not edgy; that’s called recordkeeping and indemnity management.

Should We Ban AI From Brief-Writing? No. Just supervise it.

Bans are a vibes-based solution to a supervision problem. Even before AI, our profession already punished the conduct that scares judges: hallucinated citations, false statements, and unveri fied assertions. When courts proposed AI-specific certifications, pushback wasn’t “let us hallucinate,” it was “we already certify accuracy.”

on them. Meanwhile, some judges experi mented with disclosure or certification rules, but even the Fifth Circuit (after heavy public comment) declined to adopt a special AI rule, reminding lawyers that the ordinary duties of accuracy and candor already apply. That arc should feel familiar: a burst of rulemaking, then reversion to first principles. What does “competent” AI use look like today? Boring, honestly: Treat AI like a tireless, literal intern who sometimes makes things up or makes mistakes. You can let AI summarize, brainstorm, compare, outline, and spot issues. But you must fact-check, verify citations, and keep confidences. If that sounds like supervis ing a junior associate, that’s because it is. There’s room for reasonable disagreement on copyright theory. But the Copyright Office has, for now, removed the ambiguity when it pub lished Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence on March 10, 2023. The rule is straightforward: Works that What About Copyrights— Isn’t AI “Stealing”?

The Fifth Circuit’s final word landed there. Even appellate judges are saying outright that bans are misplaced and risk withholding cost-saving tech from people who need it. Use it, own it, and move on. If a particular judge wants a certification, file it. If the applica ble rules require disclosure, disclose. But don’t confuse a moment of institutional anxiety with a long-term doctrine. The law tends to absorb technology rather than create new standards to accom modate the new tech. The Sedona Conference didn’t need the word AI to articulate the principle that discovery should focus on the needs of the case, be obtained from the most convenient and least burdensome sources, and scale with stakes. Federal Rule 26(b)(1) framed the six-factor proportionality test in 2015. AI doesn’t unsettle that; CBA RECORD 41 Discovery, Again: Proportionality Still Rules Back to where we started: relevance and proportionality.

contain purely machine-generated content aren’t registrable as authored works; only human contributions may be, and appli cants must disclose the AI portions. Courts are tracking that line. In 2025, the D.C. Circuit affirmed that an image created entirely by an AI system without human authorship isn’t protectable. Supreme Court review has since been sought in that case, Thaler v. Perlmutter , 130 F.4th 1039. You don’t need to love that result to acknowledge its clarity: Human authorship remains the gate way, and disclosure is the tollbooth. The training-data debate (i.e., is large-scale text/image inges tion fair use, licensed, or something else) is still alive in district courts, and no single opinion will “settle” it. But the right practi tioner posture isn’t nihilism; it’s risk-priced pragmatism. If your organization ships creative outputs at scale, you should (a) track how outputs are made, (b) preserve prompts, (c) know your vendors’ training and indemnity posture, and (d) keep a