AI Legal Malpractice

State of play.

• Sanctions for AI hallucinations are now a multi-jurisdictional enforcement pattern with escalating repeat-offender exposure. Judge Wang's second sanction against Kachouroff—$5,000 for a materially incorrect citation, following a $3,000 Rule 11 fine for approximately 30 defective citations in the same case—establishes that courts treat prior sanctions as aggravating, not mitigating, and will not accept human-error explanations where metadata contradicts counsel's account .
• Federal courts are extending AI accountability beyond citation errors into discovery workflow. The S.D. Indiana's White v. Walmart order holds that exclusive AI reliance in discovery review violates the FRCP good-faith meet-and-confer obligation—framing AI-driven (as distinct from AI-assisted) discovery as a sanctionable abdication of professional responsibility .
• ABA Formal Opinion 512's "reduced verification for familiar tools" standard is under direct attack. A Stanford study cited in the opinion itself documents 17–33% hallucination rates in leading legal AI platforms, and critics argue the opinion's logic collapses given that AI systems change continuously .
• Client-side AI misuse is a distinct privilege and liability vector that counsel must proactively manage. United States v. Heppner (S.D.N.Y.) held that AI-generated documents created outside counsel's direction are not privileged, and published advisory guidance now frames client AI use as a client management obligation—not merely a technology policy question .
• For counsel advising law firms, legal departments, or individual practitioners, the practical baseline is that AI governance failures are producing sanctions, resignations, and privilege waivers simultaneously across courts and jurisdictions—ethics opinion compliance is a floor, not a defense, and repeat-offender patterns are drawing escalating judicial responses.

Where things stand.

• Hallucination sanctions have a developing per-error formula. Oregon's Ringo v. Colquhoun formula ($500–$1,000 per AI error) is now being cited in federal rulings; Oregon federal courts have imposed penalties exceeding $100,000 in Green Building Initiative v. Peacock (2025); the Ghiorso case is the current appellate benchmark .
• The "no-AI policy" defense does not insulate firms from staff violations. Ghiorso had an explicit no-AI drafting policy; staff used AI anyway; the court sanctioned him regardless—establishing that policy existence without verifiable enforcement is insufficient .
• Repeat sanctions in the same case are on the record. The Kachouroff/DeMaster pattern in the Lindell litigation—two separate sanctions proceedings, contradictory excuses disproven by metadata—signals that courts treat prior sanctions as aggravating, not mitigating, factors .
• Government attorneys are not exempt. New Orleans city attorneys and a DOJ assistant U.S. attorney have resigned following sanctions proceedings over AI-generated fake citations, extending the exposure beyond private practice .
• The Seventh Circuit has admonished a former immigration judge for submitting fabricated citations in a brief, signaling that appellate courts across circuits are treating this as a disciplinary matter, not merely a procedural one .
• ABA Formal Opinion 512 / Mississippi Ethics Opinion 267 permit reduced verification for "familiar" tools—a standard critics argue is internally contradicted by the Stanford hallucination data the opinion itself cites, and which does not account for continuous model updates .
• AI-driven discovery review is categorically distinct from AI-assisted review under emerging doctrine. White v. Walmart draws the line at exclusive delegation: AI can inform attorney judgment but cannot replace the independent review required for good-faith FRCP compliance .
• Client use of consumer AI creates privilege exposure that counsel must proactively address. United States v. Heppner provides judicial backing for the proposition that uploading privileged materials to ChatGPT or Claude waives privilege; states including Pennsylvania and New York have enacted laws restricting AI impersonation of licensed professionals .
• OpenAI faces civil and criminal scrutiny over failure-to-warn obligations in the Tumbler Ridge school shooting and the Florida State University shooting—a separate but adjacent liability vector that will shape how AI companies define "imminent risk" thresholds and what disclosure obligations attach .

Latest developments.

• Judge Wang imposes a second $5,000 sanction on Kachouroff in the Lindell defamation case for a materially incorrect citation, rejecting a human-error explanation and citing precedents reaching $15,000 for fictitious citations—the second Rule 11 sanction in the same case following a $3,000 fine for approximately 30 defective citations in July 2025; Lindell and co-counsel DeMaster escaped penalty on this round, though DeMaster faced consequences in the earlier proceeding .

Active questions and open splits.

• What verification standard satisfies competence under ABA Model Rule 1.1? ABA Opinion 512 permits reduced verification for familiar tools, but the Stanford hallucination data (17–33%) and continuous model updates make "familiarity" an unstable proxy. No court has yet defined what independent verification requires in quantitative terms .
• Where is the line between AI-assisted and AI-driven work product? White v. Walmart draws it at exclusive delegation in discovery review, but the principle has not been extended to brief drafting, contract review, or due diligence workflows—each of which presents the same substitution risk .
• Does a no-AI office policy insulate the supervising attorney? Ghiorso's experience says no—but courts have not articulated what enforcement infrastructure would satisfy supervisory responsibility under Model Rule 5.1 .
• What is the privilege status of AI-assisted work product created at counsel's direction? Heppner addressed client-generated AI documents outside counsel's direction; the privilege analysis for attorney-directed AI use in drafting remains unsettled .
• Do repeat sanctions in the same case trigger bar discipline referrals? The Kachouroff/DeMaster pattern—two sanctions, contradictory excuses, metadata-disproven explanations—raises the question of whether courts will escalate to disciplinary referrals or whether monetary sanctions remain the ceiling .
• Will sanctions formulas converge across jurisdictions? Oregon's per-error formula is being cited federally, but Arizona, the Seventh Circuit, and DOJ-adjacent proceedings have not adopted a uniform metric—creating inconsistent exposure calculations for the same conduct .
• Do AI companies have a duty to warn law enforcement of threats identified in user interactions? The Tumbler Ridge and FSU shooting litigations will test whether OpenAI's internal "imminent and credible risk" threshold is legally adequate, and whether a duty to warn runs to third parties .

What to watch.

• Whether Judge Wang or other courts escalate from monetary sanctions to bar discipline referrals in repeat-offender AI citation cases—the Kachouroff pattern is the most visible test case.
• Whether any court articulates an affirmative verification protocol—specific steps, not just a standard—that satisfies Rule 1.1 competence for AI-assisted work product.
• Whether the ABA or any state bar revises the "familiar tool / reduced verification" permission in light of published criticism and the Stanford hallucination data.
• Outcome of the Tumbler Ridge victim family lawsuit and Florida AG's criminal scrutiny of OpenAI—the first cases to test whether AI companies bear a duty to warn third parties of threats identified in user sessions.
• Whether malpractice insurers begin conditioning coverage or pricing on documented AI governance protocols, creating a market-driven verification standard independent of bar guidance.
• Whether the White v. Walmart AI-in-discovery holding is adopted by other district courts or generates a circuit-level ruling on FRCP good-faith obligations.