CBA Record January-February 2026

AI 2035

GUARDRAILS FOR AI IN THE LEGAL PROFESSION How Law Schools, Lawyers, and Courts Can Use It Responsibly By Joel Bruckman

A rtificial intelligence is fundamentally reshaping how lawyers work, how law students learn, and how judges manage cases. Recognizing both the profound opportu nities and risks, the legal profession must embrace AI with dili gence and apply robust safeguards to uplift and protect our legal system. This article offers some AI guardrails for three specific seg ments of law: law schools, lawyers, and courts (see shaded boxes nearby). First, however, consider some overarching points about why guardrails matter as well as Illinois rules that all segments should be aware of. Why Guardrails Matter With AI, the magnitude of both benefit and risk is amplified. Unlike prior technologies, AI models can hallucinate or misat tribute case law. It often can’t discern subtle statutory nuances, misinterprets legal opinions, and inadvertently stores privileged or confidential information on public servers. Most concerningly, AI presents its output with polished, authoritative certainty, even when it’s completely fabricated. This occurs because, at their core, AI models rely on statistical prob abilities and are engineered to satisfy user queries—not to deliver legal truth. This is not theoretical. Illinois courts have already sanctioned lawyers for submitting briefs filled with AI hallucinations. See In

re Baby Boy, 2025 IL App (4th) 241427 (attorney who drafted brief containing largely fabricated case citations fined $1,000, ordered to return over $6,000 in fees, and referred to ARDC); In re Marla C. Martin, 24 B 13368 (N.D. Ill. Bankr. Jul. 18, 2025) (bankruptcy attorney fined $5,500 for filing brief containing AI generated cites that were fictitious or grossly misattributed).

Most concerningly, AI presents its output with polished, authoritative certainty, even when it’s completely fabricated. This occurs because, at their core, AI models rely on statisti cal probabilities and are engineered to satisfy user queries—not to deliver legal truth.

20 January/February 2026