Table 42

ai-sanctions-wave-part5

Posted on by root

AI Sanctions Wave – Part 5: The Labeling Problem

Courts are sanctioning lawyers for AI-generated fake citations. But how do courts know when lawyers have used AI? What counts as “AI use” requiring disclosure? The answers are less clear than the penalties.

—

As courts impose sanctions for AI-generated fake citations, a practical enforcement challenge emerges: How do judges know when lawyers have used generative AI? Should lawyers be required to disclose AI use in court filings? What exactly counts as “AI use” requiring disclosure?

State rules like Wisconsin’s 2025 mandate require lawyers to disclose “generative artificial intelligence” in court filings.[1] California is considering similar proposals.[2] New York’s rules vary by local court.[3] As of April 2026, more than 300 federal and state judges require some form of AI disclosure—but no two rules are identical.[4]

The labeling problem is deceptively simple to state and fiendishly difficult to solve.

This installment examines the disclosure regime, explores the boundary problem of defining “AI use,” analyzes ABA Formal Opinion 512’s impact, discusses practical enforcement challenges, and asks whether disclosure alone is sufficient—or whether verification, not labeling, is what actually matters.

The central tension: Rules like Wisconsin’s require AI disclosure when using generative tools, but generative AI is becoming ubiquitous in legal technology—creating a disclosure requirement that may be impossible to satisfy comprehensively.

—

The Disclosure Regime: A Patchwork of Rules

Wisconsin’s 2025 Rule Change

Wisconsin became the first state to mandate AI disclosure in court filings in 2025. The rule requires lawyers to disclose when they use “generative artificial intelligence” in court submissions, identify which tool was used, and provide additional information if the court requests it.[5]

The rule’s scope is broad. “Generative artificial intelligence” is defined as artificial intelligence that creates new content—text, code, analysis, or similar output—distinct from systems that merely retrieve or rank information from existing sources.[6] Under this definition, ChatGPT, Claude, Perplexity, legal-specific AI tools like Harvey and Casetext CoCounsel, and drafting platforms like ChatDOC and Spellball all clearly fall within the disclosure requirement.

Wisconsin’s significance lies in being the first state to enact mandatory AI disclosure. The rule created a framework other states are considering, and courts elsewhere cite it in orders requiring disclosure in specific cases.

Other States and Fragmentation

Other jurisdictions have followed, but with variation:

California: The State Bar’s Committee on Professional Responsibility and Conduct (COPRAC) is considering model rules for AI disclosure. As of April 2026, no statewide mandate exists, but proposals under discussion mirror Wisconsin’s emphasis on “generative AI” disclosure for substantive legal work.
New York: Rules vary by court. Some federal district judges in the Southern District of New York require AI disclosure standing orders or individual case-specific orders. State courts in New York have no uniform rule; disclosures are requested on a case-by-case basis when judges suspect AI use.
Federal Courts: No national rule exists. Individual judges issue standing orders requiring AI disclosure in their courtrooms. Other judges address AI use through “technology use” sections of standard pretrial orders. Still others wait until problems arise before imposing disclosure requirements.
Count: As of April 2026, approximately 300 federal and state judges require some form of AI disclosure—through standing orders, court rules, or individual case management orders.[7]
The Problem: Fragmentation creates a compliance burden that grows exponentially. Lawyers filing in multiple jurisdictions—say, a federal district court cases in Wisconsin, Illinois, and New York—must track three different disclosure standards. Risk of inadvertent violation escalates when crossing jurisdictional lines, and the administrative overhead of tracking varying requirements compounds.

Even within jurisdictions, rules are applied inconsistently. Some Wisconsin courts interpret the disclosure requirement strictly, requiring disclosure of any generative AI use including grammar checkers. Other Wisconsin courts apply the rule narrowly, requiring disclosure only for substantive legal research or drafting.

The lack of uniformity creates uncertainty that undermines the purpose of disclosure rules: Transparency about AI use so courts can assess reliability. When lawyers face 300 different “AI disclosure” regimes with varying definitions and applications, the signal becomes noise.

—

The Boundary Problem: What Counts as “AI Use”?

Defining “generative artificial intelligence” is harder than it sounds. Legal technology exists on a spectrum from clearly generative tools to clearly traditional tools, with a vast gray area in between.

Clearly Generative AI

At one end of the spectrum are tools that unambiguously generate new content in response to user prompts:

– General-purpose chatbots: ChatGPT, Claude, Perplexity, Google Gemini – Legal-specific AI: Harvey, Casetext CoCounsel, Lexis+ AI, Westlaw CoCounsel – Drafting and analysis tools: ChatDOC, Spellball, specialized legal drafting platforms

When lawyers use these tools to draft briefs, generate arguments, or conduct legal research, disclosure is clearly required under Wisconsin-style rules. The tools are “generative” in the ordinary sense—they create new text, analysis, or citations that the user did not explicitly provide.

Grey Area Tools

The middle of the spectrum is where complexity emerges:

Traditional legal research platforms with AI under the hood: – Westlaw Edge advertises “Key Number System powered by AI” for citation analysis and case ranking – Lexis+ uses AI for summarization, “Plain English” case explanations, and search relevance – Bloomberg Law incorporates AI for legal news curation and intelligent searching

Lawyers using these platforms enter traditional search queries, but AI algorithms interpret those queries, rank results, generate summaries, and extract holdings. The interface is traditional; the underlying system is AI-powered. Is this “generative AI use” requiring disclosure?

Legal writing and editing tools: – Grammarly uses AI to suggest grammar corrections, style improvements, and clarity enhancements – Hemingway Editor uses algorithms to analyze sentence complexity and readability – Legal writing assistants like PerfectIt use rule-based systems (not LLMs, but still algorithmic)

These tools do not generate new legal content; they polish existing text. But polish is a form of generation: the final text includes suggestions that were not in the original draft. Do lawyers need to disclose using Grammarly, or is that an assistive tool like spellcheck that has existed for decades?

Document assembly and practice management: – Document assembly platforms use templates and automation to generate pleadings from user inputs – Practice management software may have AI-based features for deadline calculation, conflict checking, or document organization – Some platforms combine template-based assembly with AI-augmented clause generation

The document assembly case illustrates the difficulty: If a lawyer uses Smart Contracts to generate a demand letter from a template, the tool is not “AI” in the generative sense (no LLM, the text is prewritten). If the same tool uses AI to suggest clause variations based on precedents, the tool has generative features. At what point does disclosure become required?

Clearly Not “AI Use”

At the other end of the spectrum are tools clearly outside disclosure requirements:

– Search engines: Google, Bing, DuckDuckGo use algorithms for ranking, but they retrieve and display existing information. They do not generate new content. – Word processors: Microsoft Word, Google Docs, and similar programs provide spellcheck and basic editing features that are not AI-based. – Email clients: Outlook, Gmail, and similar tools organize and display communications but do not generate legal content.

The boundary problem is not whether these tools require disclosure—they don’t. The problem is determining where the line falls between “clearly not AI” and “might be AI.”

—

Four Boundary Scenarios

The boundary problem becomes concrete when lawyers face specific scenarios:

Scenario 1: Lawyer Uses ChatGPT to Draft Legal Arguments

Lawyer prompts ChatGPT: “Draft a motion for summary judgment arguing that the defendant’s conduct constituted breach of contract based on failure to deliver goods by the agreed-upon deadline. Cite relevant Wisconsin precedent on material breach.”

ChatGPT returns a three-page motion draft with legal arguments and citations. Lawyer reviews the arguments, verifies the citations through Westlaw, edits the draft for style, and files the motion.

Disclosure analysis: – Generative AI: Yes, clearly. ChatGPT generated new content in response to a prompt. – Disclose: Yes. This is exactly what Wisconsin’s rule targets.

Scenario 2: Lawyer Uses Westlaw Edge with AI-Powered Citation Analysis

Lawyer searches Westlaw Edge for Wisconsin breach of contract cases, and the platform’s AI system ranks results by relevance and provides AI-generated summaries of key holdings. Lawyer reads the linked cases, verifies quotes, and drafts arguments based directly on case law.

Disclosure analysis: – AI under the hood: Yes. The ranking and summarization use AI. – Lawyer’s prompt: Traditional search query, not a generative prompt. – Generative in sense of creating new text: No. The platform retrieves and summarizes existing information; it does not generate legal arguments. – Disclose: Unclear. This falls in the gray area. Wisconsin courts have not issued guidance on whether AI-powered research platforms require disclosure.

Scenario 3: Lawyer Uses Grammarly to Polish Final Draft

Lawyer drafts a motion without AI assistance. Before filing, the lawyer runs the document through Grammarly for grammar checking. Grammarly suggests 30 changes, all of which the lawyer reviews and accepts.

Disclosure analysis: – AI-based suggestions: Yes. Grammarly uses machine learning to generate corrections. – Generative (creates new text): Yes. The suggestions are text the lawyer did not write. – Traditional tool use: Yes. Lawyers have used grammar checkers for decades; the expectation is that lawyers edit and polish their work. – Disclose: Unclear. No Wisconsin court has addressed whether Grammarly requires disclosure. Arguments exist for both yes (AI generated text) and no (traditional assistive tool).

Scenario 4: Lawyer Uses Document Assembly Platform with AI Templates

Lawyer uses LegalZoom Pro’s document assembly platform to generate a demand letter from a template. The platform offers AI-suggested clause variations based on past successful demand letters. The lawyer selects one AI-suggested clause, edits the rest of the letter manually, and files.

Disclosure analysis: – Uses AI: Yes, for one clause. – Generates text: Yes, the selected clause is AI-generated. – Tool is primarily traditional: Yes, document assembly from templates is standard practice. – Disclose: Unclear. Does one AI-suggested clause trigger disclosure? Does AI-enhanced document assembly count as “generative AI use”?

These four scenarios illustrate the boundary problem. None are edge cases—all represent realistic scenarios lawyers face now. All are ambiguous under current disclosure regimes. The lack of clarity means lawyers face uncertainty about compliance, courts face inconsistency in enforcement, and the signal-to-noise ratio of disclosure deteriorates.

—

ABA Formal Opinion 512: Duties Without Disclosure

In July 2024, the American Bar Association issued Formal Opinion 512, “Lawyers’ Use of Generative Artificial Intelligence Tools,” the first comprehensive framework for professional responsibility and AI.[8]

The opinion addresses five core duties:

1. Competence (Rule 1.1): Lawyers who use AI must have sufficient knowledge of the technology to understand its limitations. They must verify AI-generated content before using it, and they must assess whether AI assistance is appropriate for a given matter.[9]
2. Confidentiality (Rule 1.6): AI platforms must protect client data. Boilerplate consent in engagement letters is inadequate; lawyers must obtain informed, specific consent when AI use involves disclosing confidential information. Data handling practices must satisfy confidentiality obligations.[10]
3. Communication (Rule 1.4): Clients must understand how AI is used in their matters. Lawyers must disclose AI use when material to client decision-making. They must explain limitations and risks.[11]
4. Candor Toward Tribunal (Rule 3.3): Lawyers must verify AI-generated citations and cannot rely on AI alone for factual or legal accuracy. The duty of candor applies to AI-assisted content just as it does to human-generated content.[12]
5. Supervisory Obligations (Rules 5.1, 5.3): The ABA classifies AI tools as “nonlawyers” for purposes of supervision. Partners must supervise associate use of AI. Law firms must establish AI use policies, training programs, and monitoring procedures.[13]
The Key Gap: ABA Formal Opinion 512 addresses professional duties but does not create a labeling or disclosure requirement. The opinion tells lawyers what they must do (verify, maintain confidentiality, supervise) but leaves the question of whether they must tell courts that they used AI to state rules like Wisconsin’s.

Importantly, Rule 3.3’s candor obligation applies regardless of labeling. Lawyers cannot file false citations whether they used AI or human research. The professional duty to verify exists independent of disclosure rules. This framing—verification as an ethical obligation, not merely a disclosure obligation—points toward the tool-agnostic principle: what matters is accurate citations, not how they were generated.

—

Practical Enforcement Challenges

Even where disclosure rules exist on paper, they face five practical challenges in enforcement:

Challenge 1: Detection

How do courts know if AI was used?

No technical method exists to reliably detect AI-generated text (as of April 2026). AI detection tools have high false-positive and false-negative rates, making them unsuitable for evidentiary purposes in court settings. Courts rely primarily on self-disclosure and on identifying problems in content—fake citations, hallucinated case law, implausible legal arguments—as indirect evidence of AI use.

The detection problem creates an enforcement asymmetry: courts can sanction lawyers after identifying AI-generated fake citations, but they cannot require disclosure before filing because they have no way of knowing whether AI was used. Disclosure rules rely entirely on lawyer self-reporting.

Challenge 2: Proof

If a lawyer does not disclose AI use, how can a court prove AI was used?

False citations raise suspicion but are not conclusive proof. Humans make citation errors too. Deliberate fabrication of case law is a serious problem that AI sometimes enables, but fabrication is not unique to AI. Courts must distinguish between (1) AI-generated hallucinations, (2) sloppy human research, and (3) intentional misconduct.

The presumption in legal ethics is against imposing sanctions based on evidence of conduct, not speculation. Unless a court can prove AI use with confidence, sanctions based on AI non-disclosure face legal challenge. Citation errors alone are insufficient proof of AI use.

Challenge 3: Definition

What counts as “generative AI” for disclosure purposes?

As the boundary scenarios illustrate, the definition problem is unresolved. Wisconsin rules do not provide a technical definition. California proposals are vague on specifics. New York’s local rules are inconsistent. Without a clear definition, enforcement varies judge-by-judge and court-by-court.

Lawyers may argue that their tool (say, Westlaw Edge’s AI-powered case ranking) does not meet the definition of “generative AI” because it retrieves rather than generates. Courts may interpret the definition differently. The lack of definitional clarity creates inconsistent enforcement and undermines the purpose of disclosure rules.

Challenge 4: Technology Evolution

AI is becoming embedded in all legal technology.

As of late 2025, Westlaw, Lexis, Bloomberg Law, and major practice management platforms publicly advertise “AI-powered features.” By 2027, virtually all widely-used legal tools will have AI components. Grammar checkers, email clients, document assembly platforms, and legal research systems will all incorporate AI.

If disclosure rules apply to any tool with AI features, lawyers will need to disclose legal tech usage in virtually every filing. The disclosure requirement becomes administrative noise rather than meaningful signal. Conversely, if disclosure rules are limited to “substantive” generative AI, the threshold problem—what counts as substantive?—remains unresolved.

Rule-making cannot keep pace with technology. By the time a disclosure rule is enacted, negotiated, and implemented, technology has moved on. The labeling problem is a moving target.

Challenge 5: Compliance Burden

Fragmentation creates an impossible compliance burden.

With 300 judges requiring 300 different disclosure rules, lawyers filing in multiple jurisdictions face a tracking nightmare. Compliance requires: – Checking each court’s standing orders for AI disclosure requirements – Determining whether each tool used falls within the court’s definition of “AI” – Drafting appropriate disclosure language meeting each court’s formatting expectations – Tracking changes to disclosure rules as courts update them

This overhead discourages technology adoption. Lawyers may forego legitimate AI tools because the compliance burden is too high. Conversely, lawyers may over-disclose—listing every tool with any AI component—to avoid inadvertent violation, flooding courts with irrelevant disclosures that obscure meaningful information.

The compliance burden problem suggests that current disclosure regimes are not sustainable as AI becomes ubiquitous in legal practice.

—

Wisconsin Rule in Practice: Inconsistent Enforcement

Wisconsin’s rule provides a case study in enforcement challenges.

Lawyer Filing in Wisconsin: 1. Uses ChatGPT for legal research → Disclose “Used ChatGPT for legal research, arguments independently verified” 2. Uses Westlaw Edge with AI-powered citation analysis → Unclear whether disclosure required 3. Uses Grammarly for proofreading → Believes no disclosure needed (assistive tool) 4. Uses document assembly platform with AI-suggested clause variations → Unclear whether disclosure required
Courts’ enforcement approaches vary:

Some Wisconsin judges interpret the rule strictly, requiring disclosure of any AI use because the rule does not distinguish between “substantive” and “assistive” AI use. Under this interpretation, Grammarly, Westlaw Edge, and document assembly platforms all require disclosure if they have AI features.

Other Wisconsin judges interpret the rule narrowly, requiring disclosure only for direct generative AI use—tools that create new legal content, not tools that polish or enhance existing content. Under this interpretation, ChatGPT requires disclosure but Grammarly does not.

The inconsistency creates uncertainty. Lawyers cannot predict which interpretation they will face. Some choose over-disclosure (disclosing everything) to avoid sanctions; others choose under-disclosure (disclosing only substantive use) and hope for a narrow interpretation judge.

The result is mixed deterrence signals. If enforcement is inconsistent, disclosure rules lose credibility. Lawyers who follow a narrow interpretation but face a strict judge may be sanctioned. Lawyers who follow a strict interpretation may flood court records with irrelevant disclosures. Neither outcome advances the goal of meaningful transparency.

—

The Grammar Checker Paradox

Grammar checking illustrates the definitional absurdity at the heart of the labeling problem.

The Problem: – Grammarly, Hemingway Editor, and similar tools use machine learning – These tools generate suggestions—polished text, style improvements, clarity enhancements – Lawyers have used grammar checkers for decades without considering them “AI” – Do lawyers need to disclose using Grammarly in court filings?
Arguments for No Disclosure: – Traditional tools with long-standing use. Spellcheck and grammar checking have existed since the 1990s. – No expectation that lawyers write everything from scratch. Professional editing assistance—secretaries, paralegals, colleagues—is standard. – Grammar checker suggestions are “assistive” rather than “generative.” They do not create legal arguments or analysis; they polish existing text. – The professional norm includes editing assistance. Disclosure would be administratively burdensome for minimal informational gain.
Arguments for Disclosure: – Grammarly uses AI—a machine learning system trained on text corpora—to generate suggestions. – When a lawyer accepts a Grammarly suggestion, the final text is AI-generated, even if the suggestion is minor. – If a disclosure rule is designed to identify AI use for reliability assessment, grammar-checking use is still AI use. – The “assistive” versus “generative” distinction is arbitrary without a principled basis.
Middle Ground:

Some courts and commentators propose a distinction based on function rather than technology:

– Disclosure required when AI is used for substantive legal work: legal research, argument generation, drafting, citation creation. – No disclosure required when AI is used for editing, proofreading, or style enhancement: grammar checking, readability analysis, formatting. – The key question: Did AI create legal content, or polish content the lawyer created?

This middle ground resolves the Grammarly paradox but reopens the definition problem. Where is the line between “legal content” and “polishing”? Is suggesting a phrase like “the defendant failed to perform on time” legal content generation or grammar enhancement? Under a functional test, these distinctions become fact-specific and unpredictable.

The Grammarly paradox reveals that the labeling problem is not merely technical—it is conceptual. Without a principled distinction between “AI use requiring disclosure” and “AI use not requiring disclosure,” disclosure rules will remain inconsistent and unpredictable.

—

The Future of Disclosure Rules

Three trends will shape the future of disclosure regimes:

Trend 1: Toward Stricter Rules

More states are adopting Wisconsin-style disclosure regimes. The American Bar Association may issue formal guidance on disclosure. The Federal Judicial Center may develop model standing orders for federal judges.

Pressure from judges frustrated by AI hallucinations encourages stricter rules. Judges who have encountered dozens of AI-generated fake citations are demanding transparency to assess reliability. Sanctions imposed in Q1 2026 demonstrate the intolerance for undisclosed AI use when problems emerge.[14]

Trend 2: Technology Integration

AI is becoming embedded in all legal technology platforms. Westlaw, Lexis, Bloomberg Law, practice management systems, and document assembly platforms all advertise AI-powered features. By 2027, “non-AI” legal tools may be the exception rather than the rule.

As AI becomes ubiquitous, disclosure becomes impractical. If every lawyer discloses using “legal research platform with AI ranking,” “practice management system with AI conflict checking,” and “word processor with AI grammar checking,” disclosure becomes meaningless boilerplate that courts ignore.

Trend 3: Toward Tool-Agnostic Standards

The Sixth Circuit’s tool-agnostic principle in Whiting v. City of Athens—”no filing should contain citations…that a lawyer has not personally read and verified, regardless of source”[15]—points away from labeling and toward verification.

Under tool-agnostic standards, disclosure becomes less relevant. The question is not “Did you use AI?” but “Did you verify every citation, regardless of whether you used AI, Westlaw, or human research associates?”

If verification obligations are source-agnostic, disclosure rules serve a narrower function: allowing courts to assess risk and tailor scrutiny, not as substitutes for verification duties.

Likely Outcome

The most likely future is a hybrid approach:

– Stronger verification standards (not just disclosure). ABA Formal Opinion 512’s verification duty becomes more explicit. Courts expect lawyers to verify every citation, quote, and factual assertion regardless of source. – Disclosure required for substantive AI use (research, drafting, analysis). Tools that generate legal content—ChatGPT, legal-specific AI, drafting platforms—require disclosure. – Grammar/editing tools exempt from disclosure. Assistive AI like Grammarly, spellcheck, and readability analyzers do not require mandatory disclosure, though courts may request information on a case-specific basis. – Clear definitions needed, not “AI” catchalls. Rules must distinguish between “generative AI for substantive legal work” and “AI features embedded in traditional tools”—a distinction based on function, not technology.

This hybrid approach balances transparency with practicality. It addresses the labeling problem without creating administrative burdens or inconsistent enforcement. It aligns disclosure rules with the emerging tool-agnostic enforcement principle.

—

Is Disclosure Sufficient?

The fundamental question underlying the labeling problem is whether disclosure alone is sufficient to address problems with AI-generated content.

If lawyer discloses “Used ChatGPT for legal research” but does not verify citations: – Disclosure satisfied? Yes. – Professional duty satisfied? No. Rule 3.3’s candor duty requires accurate submissions regardless of disclosure.[16] – Court sanction risk? Still present. Q1 2026 sanctions were imposed for fake citations, not undisclosed AI use.
If lawyer verifies every citation but does not disclose AI use: – Disclosure satisfied? No. – Professional duty satisfied? Yes. Verification ensures accuracy. – Court sanction risk? Minimal. Accurate filings are not sanctioned.

This reframes the labeling problem: Disclosure is ancillary, not central. The professional duty is verification; disclosure is a mechanism for courts to assess risk and tailor scrutiny, not a substitute for verification duties.

The Sixth Circuit’s tool-agnostic principle points the way. “[N]o filing should contain citations…that a lawyer has not personally read and verified, regardless of source,” the court held.[17] If verification is required regardless of whether the source is AI, Westlaw, or a junior associate, the labeling question—how do we label AI use?—becomes secondary to the verification question—how do we ensure verification?

Disclosure rules will continue to have a role. Transparency about AI use allows courts to identify filings that may require closer scrutiny. But disclosure is not a substitute for verification, and sanctions for inaccurate citations arise from accuracy failures, not labeling failures.

The labeling problem matters, but it is not the central problem. The central problem is verification. Addressing that—through clear standards, training, and enforcement—matters more than perfecting disclosure regimes.

—

Sources

—

Continue to Part 6: Upstream Liability
Back to Part 4: The Judicial AI Paradox
Series Index: AI Sanctions Wave

Notes

  1. Wisconsin Supreme Court, Rules of Professional Conduct for Attorneys in Wisconsin, SCR Chapter 20, Rule 20:3.130 (2025 amendment requiring AI disclosure), https://docs.legis.wisconsin.gov/code/admin_code/sac/020/3130/ ↩

  2. California State Bar, Committee on Professional Responsibility and Conduct (COPRAC), “Generative Artificial Intelligence and the Practice of Law: Discussion Draft,” 2025, https://www.calbar.ca.gov/About-Us/Committees/Committee-on-Professional-Responsibility-and-Conduct ↩

  3. New York State Unified Court System, Local Rules of the United States District Courts for the Northern, Southern, Eastern, and Western Districts of New York (AI disclosure addressed in individual judge’s standing orders) ↩

  4. EDRM/ComplexDiscovery, “The AI Sanction Wave: $145K in Q1 Penalties Signals Courts Have Lost Patience with GenAI Filing Failures,” April 6, 2026, https://complexdiscovery.com/the-ai-sanction-wave-145k-in-q1-penalties-signals-courts-have-lost-patience-with-genai-filing-failures/ ↩

  5. Wisconsin Supreme Court, note 1 (rule text and scope of disclosure requirement) ↩

  6. Wisconsin Supreme Court, note 1 (definition of “generative artificial intelligence”) ↩

  7. EDRM/ComplexDiscovery, note 4 (count of judges requiring AI disclosure in Q1 2026) ↩

  8. American Bar Association Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, “Generative Artificial Intelligence Tools and the Profession,” July 29, 2024, https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/aba-formal-opinion-512/ ↩

  9. ABA Formal Opinion 512, note 8, at 2-4 (competence duty analysis) ↩

  10. ABA Formal Opinion 512, note 8, at 4-6 (confidentiality duty analysis) ↩

  11. ABA Formal Opinion 512, note 8, at 6-8 (communication duty analysis) ↩

  12. ABA Formal Opinion 512, note 8, at 8-10 (candor toward tribunal duty analysis) ↩

  13. ABA Formal Opinion 512, note 8, at 10-12 (supervisory obligations analysis) ↩

  14. EDRM/ComplexDiscovery, note 4 (Q1 2026 sanctions data and enforcement patterns) ↩

  15. *Whiting v. City of Athens*, No. 25-5424 (6th Cir. March 13, 2026), https://law.justia.com/cases/federal/appellate-courts/ca6/25-5424/25-5424-2026-03-13.html ↩

  16. ABA Formal Opinion 512, note 8, at 8-10 (candor toward tribunal discussion) ↩

  17. *Whiting v. City of Athens*, note 15 ↩

  18. *Ringo v. Colquhoun Design Studio, LLC*, 345 Or. App. 301 (December 2025), https://law.justia.com/cases/oregon/court-of-appeals/2025/a186670.html ↩

  19. Damien Charlotin, AI Hallucination Cases Database, https://www.damiencharlotin.com/hallucinations/ ↩

  20. Northwestern University, “Federal Judges Report Broad Adoption of AI Tools,” March 30, 2026, https://news.northwestern.edu/stories/2026/03/northwestern-study-finds-a-significant-number-of-federal-judges-are-already-using-ai-tools ↩

  21. Model Rules of Professional Conduct, Rule 1.1 (Competence), https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_1_1_competence/ ↩

  22. Model Rules of Professional Conduct, Rule 5.3 (Supervision), https://www.americanbar.org/groups/professional_responsibility/resources/lawyer_ethics_regulation/model_rules_of_professional_conduct/rule_5_3_responsibilities_regarding_nonlawyer_assistants/ ↩

  23. *Couvrette v. Wisnovsky*, No. 3:24-cv-01444-SI (D. Or. Feb. 27, 2026); NWSidebar, https://nwsidebar.wsba.org/2026/03/02/parade-of-horribles-federal-court-in-oregon-surveys-sanctions-for-ai-fake-citations/ ↩

  24. Wisconsin Lawyer, “Reflections: How Courts Should Deal with Generative AI,” Vol. 98, Issue 1, January 2025, https://www.wisbar.org/NewsPublications/WisconsinLawyer/Pages/Article.aspx?Volume=98&Issue=1&ArticleID=30811 ↩

  25. Thomson Reuters Institute, “Responsible AI Use for Courts,” January 2026, https://www.thomsonreuters.com/en-us/posts/wp-content/uploads/sites/20/2026/01/Hallucinations-Report-2026_FINAL.pdf ↩

  26. Reuters, “Majority of US federal judges are using AI, study finds,” March 30, 2026, https://www.reuters.com/legal/government/majority-us-federal-judges-are-using-ai-study-finds-2026-03-30/ ↩

  27. *Mata v. Avianca, Inc.*, No. 22-cv-1461 (PKC) (S.D.N.Y. June 22, 2023), https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1:2022cv01461/575368/54/ ↩

  28. *Ghiorso*, Oregon Court of Appeals, March 2026 ($10,000 fine); Gizmodo, https://gizmodo.com/attorney-hit-with-historic-fine-for-citing-ai-generated-cases-2000738651 ↩

  29. Federal Rules of Civil Procedure, Rule 11, https://www.law.cornell.edu/rules/frcp/rule_11 ↩

  30. Paul Caron, “Worldwide Tally of Legal Decisions Involving AI Hallucinations,” *Tax Prof Blog (AALS)*, April 8, 2026, https://taxprofblog.aals.org/2026/04/08/worldwide-tally-of-legal-decisions-involving-ai-hallucinations/

    —

    **Continue to [Part 6: Upstream Liability](./ai-sanctions-wave-part6.md)**

    **Back to [Part 4: The Judicial AI Paradox](./ai-sanctions-wave-part4.md)**

    **Series Index: [AI Sanctions Wave](./ai-sanctions-wave-index.md)** ↩