AI Litigation Briefing — September 15, 2026
Updated: Sep 26
This week's highest-ranked Litigation AI GRC developments concentrate on active disputes and concrete governance failures: the OpenAI-New York Times fair-use fight, a New Mexico contempt sanction over AI-fabricated testimony, Quebec judicial GenAI guidelines, Florida's OpenAI consumer-protection forum dispute, an Oklahoma judge's fictitious citations, two federal discovery orders involving AI workflows and search terms, xAI's Minnesota constitutional challenge, expert prompt discovery, and a major law-firm AI investment. The common operational issue is reconstructability: can counsel identify authentic sources, the AI workflow, governing legal status, human decision-maker, preserved records and unresolved uncertainty when challenged? This briefing keeps complaints separate from findings, discovery orders separate from merits rulings, guidance separate from law, and technology claims separate from independent assurance. It is neutral Litigation AI GRC analysis for North American legal decision-makers, not legal advice.
This is an honest AI disclosure. This briefing presents my analysis, as Pouya Shafabakhsh, from the perspective of AI governance, risk, compliance, and AI litigation. For the convenience of legal professionals and busy executives, this page also includes a podcast. The written briefing on this page is the controlling substantive analysis; a verbatim transcript is not currently available.
I. OpenAI-New York Times Fair-Use Fight Reaches Summary-Judgment Stage
Case or Development Summary and Status
Reuters reported that OpenAI, Microsoft, The New York Times and groups of authors have moved a central question in the consolidated Manhattan copyright litigation toward summary judgment: whether unauthorized use of copyrighted books and news articles to train generative AI systems qualifies as fair use. The publishers and authors argue that the training use exploits protected expression and can impair licensing or substitution markets. OpenAI and Microsoft argue that training is transformative because models analyze works to learn statistical relationships rather than distribute the source works as such. The dispute is part of In re OpenAI Inc. Copyright Infringement Litigation, No. 1:25-md-03143 in the Southern District of New York. The motions ask the court to decide legal issues on a developed record, but the filing of cross-motions is not itself a finding for either side. On September 14, Reuters separately reported that the magistrate judge handling pretrial matters would remain on the case after divesting Microsoft stock. That procedural development does not resolve the fair-use merits.
Litigation AI GRC Analysis
For Litigation AI GRC, the immediate issue is evidence architecture. Training-data provenance, acquisition methods, licensing positions, model-development records, output testing, internal legal analysis, and market-impact evidence can all become consequential. Organizations should distinguish three separate questions: whether training copies were lawfully obtained, whether the training use is legally permissible in the relevant jurisdiction, and whether model outputs create additional infringement or market-substitution issues. California decisions involving other developers may inform arguments, but they do not automatically decide this SDNY litigation. The control objective is to make the factual chain reconstructable without converting unsettled doctrine into a universal rule.
Strategic Risk Management Recommendation
AI developers, publishers, licensors and enterprises fine-tuning models should maintain proportionate records of source acquisition, license terms, dataset restrictions, model or dataset changes, rights-holder notices, and material output testing. Where third-party models or datasets are used, contracts should identify data-use rights, cooperation duties, preservation responsibilities and access to records needed for disputes. Once litigation is reasonably anticipated, counsel should direct preservation of relevant model versions, datasets, communications and testing artifacts. Keep legal conclusions separate from technical observations and do not describe pending fair-use arguments as settled law.
II. New Mexico Supreme Court Fines Lawyer Over AI-Fabricated Testimony
Case or Development Summary and Status
Aarons had used the system to summarize trial proceedings and failed to verify the resulting material against the record. The court referred him to the disciplinary board for further review and reassigned the appeal to a public defender. The ruling is significant because the defect involved invented record evidence, not merely a nonexistent case citation. The sanction does not mean every use of generative AI in appellate work is prohibited. It reflects the consequences of submitting unverified machine-generated material in a filing where counsel remains responsible for accuracy and the client's representation. Any later disciplinary outcome should be tracked separately from the contempt order already reported.
Litigation AI GRC Analysis
For Litigation AI GRC, this is a direct example of why source authenticity must be controlled at the point of reliance. A system that summarizes a transcript can create plausible but nonexistent testimony, names or events if it is not grounded to the actual record. The risk is especially acute in appeals, dispositive motions, expert work and factual chronologies because a reviewer may focus on legal argument and miss fabricated factual detail. Governance should distinguish AI-assisted organization from evidentiary truth. Human review is meaningful only if the reviewer checks the machine output against authoritative source material rather than reviewing the generated text in isolation.
Strategic Risk Management Recommendation
Firms should require record-based propositions in consequential filings to be traceable to the official transcript, exhibit, docket entry or other authoritative source. For AI-assisted summaries, use a verification workflow that preserves source references, reviewer identity, material corrections and final sign-off. Prohibit insertion of unverified quotations, testimony, names or citations into filed work product. Training should make clear that a fluent answer is not evidence. If a material hallucination is discovered after filing, escalate promptly to responsible counsel for matter-specific corrective action rather than silently editing internal copies.
III. Quebec Courts Adopt Common GenAI Guidelines for Judges
Case or Development Summary and Status
The Quebec Court of Appeal announced that the Court of Appeal, Superior Court, Court of Quebec and Municipal Courts adopted common guidelines for judges' use of generative AI. The official notice says the framework is particularly relevant where an authorized, secure institutional tool has not been deployed within a court's operational environment. The guidelines preserve the human character of judging and address the security, confidentiality and reliability concerns that arise when judges consider generative tools. Public reporting on the framework states that judicial decision-making cannot be delegated, in whole or in part, to a generative AI system. Permitted uses are bounded and may include administrative or descriptive assistance rather than substitution for the judicial function. This is a judicial-governance development, not a statute and not a blanket rule governing every Canadian court. Ontario, federal and other provincial courts must be assessed under their own authorities and practices.
Litigation AI GRC Analysis
For Litigation AI GRC, the Quebec development matters beyond judges because it demonstrates a governance pattern relevant to any high-stakes legal decision: approved tools, secure environments, confidentiality, independent verification, non-delegation of core judgment and visible responsibility for the final decision. Counsel appearing before courts should not assume that judicial adoption of a technology changes counsel's own professional duties or creates a universal disclosure requirement. Equally, firms should expect courts to become more sophisticated about asking how AI-assisted material was created and verified where reliability becomes material to a proceeding.
Strategic Risk Management Recommendation
Organizations supporting litigation should separate low-risk administrative assistance from functions that affect evidence, legal judgment, credibility, findings or outcomes. Maintain an approved-tool register, restrictions for confidential information, a human-review checkpoint and an escalation path for uncertain outputs. For court-facing work, identify the actual court-specific AI notices or directions before making any disclosure statement. Do not generalize Quebec guidance to Ontario, New York or U.S. federal courts. Where a secure institutional tool is unavailable, use more restrictive controls for confidential or privileged material.
IV. Florida's OpenAI Consumer-Protection Case Returns to State Court
Case or Development Summary and Status
The Florida Bar's September 10 news summary reports that U.S. District Judge Aileen Cannon agreed with Florida's attorney general that a consumer-protection case against OpenAI and CEO Sam Altman belongs in state court rather than federal court. The order remanded the matter to Florida's 10th Judicial Circuit, where Judge Heather Beato is presiding. According to the report, the federal court rejected the argument that references tracking federal standards under the Children's Online Privacy Protection Act were enough to create federal-question jurisdiction under the Grable doctrine. The ruling is procedural: it determines the forum, not whether Florida's substantive allegations are correct. The state's complaint remains contested, and liability, remedies and factual findings will depend on the state-court litigation. The development is nevertheless important because state consumer-protection theories are becoming a distinct lane of AI litigation alongside copyright, privacy, product liability and professional responsibility.
Litigation AI GRC Analysis
For Litigation AI GRC, forum shifts can change discovery practice, scheduling, available claims, state-law standards and the decision-makers who will evaluate platform conduct. Companies operating nationally should not treat federal AI policy as the only source of exposure. State attorneys general can pursue traditional consumer-protection statutes against allegedly deceptive, unsafe or unfair technology practices. Governance evidence may therefore include marketing claims, age-related controls, safety representations, internal testing, complaint handling and communications with regulators. The remand also illustrates why a legal-status register should distinguish procedural holdings from findings on the underlying AI behavior.
Strategic Risk Management Recommendation
Enterprises offering AI products to consumers should map public claims about safety, privacy, age controls and performance to the controls that actually exist. Preserve versioned policies and material testing, and ensure complaint and incident records can be reconstructed without collecting unnecessary personal data. Litigation teams should maintain a forum-specific matrix for state and federal claims rather than assuming one national rule. When a case is remanded, update preservation, discovery and reporting plans to the receiving court and applicable state procedure. Do not restate allegations as proven facts.
V. Oklahoma Judge's AI-Assisted Order Included Fictitious Cases
Case or Development Summary and Status
State records prompted an Oklahoma State Bureau of Investigation inquiry, and the attorney general declined a recommendation for criminal prosecution. It should be described carefully: an investigation or referral is not a disciplinary finding, and the decision not to prosecute does not resolve every question about judicial administration or professional standards. The immediate verified point is narrower - a court order contained nonexistent authorities after the judge used an AI tool. The incident is operationally important because judicial AI errors can affect litigants even when counsel themselves did not use the technology.
Litigation AI GRC Analysis
For Litigation AI GRC, the event expands the verification problem from law-firm filings to the decision layer. Litigants need processes for identifying material citation or factual anomalies in orders, expert reports and opposing submissions without treating every mistake as evidence of AI use. Courts likewise need secure research workflows and independent source verification. A citation checker can help, but it is not sufficient if a real case is cited for a proposition it does not support. The defensible control is authoritative-source review paired with human responsibility for the resulting judicial or legal work.
Strategic Risk Management Recommendation
Legal teams should verify consequential authorities in orders and briefs against official or reliable case-law sources when an issue could affect rights, deadlines or appellate strategy. Internally, prohibit reliance on generated citations without opening and reading the underlying authority. Track material AI errors as incidents and use them to improve training and controls, while avoiding unsupported accusations about how another participant produced an error. Courts and firms adopting AI research tools should test them against known false citations, quotation alterations and jurisdiction-specific authorities before high-stakes deployment.
VI. Elgin v. Wells Fargo: Court Rejects Claimed AI Protective-Order Violation
Case or Development Summary and Status
In Elgin v. Wells Fargo Bank, N.A., No. 4:25-cv-03937-KAW, Magistrate Judge Kandis A. Westmore addressed three discovery disputes in which Wells Fargo challenged the plaintiff's use of artificial intelligence. The order records Wells Fargo's contention that the plaintiff uploaded discovery requests into an AI tool and then responded to rewritten requests. The plaintiff said the uploaded material was not confidential, that AI was not used to answer the rewritten requests, and that Wells Fargo had not yet produced documents marked confidential at the relevant time. The court found the plaintiff's points persuasive, declined to find a violation of the stipulated protective order, and rejected the extreme relief sought on that theory. The court nonetheless required supplemental responses to a request for production and an interrogatory within 14 days. The ruling is fact-specific. It does not create a general safe harbor for uploading discovery material to AI systems.
Litigation AI GRC Analysis
For Litigation AI GRC, the order is valuable because it shows how AI use can become a discovery dispute even before confidential production is at issue. The governing questions remain concrete: what information entered the tool, whether it was protected, what the tool did to the request, whether the resulting response accurately answered the propounded discovery, and whether the system's contractual and retention settings create separate confidentiality risks. Counsel should avoid collapsing those issues into a single statement that 'AI was used.' The legal consequence depends on the protective order, actual data, workflow and prejudice.
Strategic Risk Management Recommendation
Before using AI in discovery, define approved tasks and prohibited data classes. Preserve the original discovery request, source materials, AI-assisted draft where material, human edits and final response so the chain can be reconstructed if challenged. Review protective orders and confidentiality designations before uploading material, and separately review vendor data-use and retention terms. When AI rewrites or summarizes a discovery request, require the responding lawyer to compare the final answer to the original request, not merely the machine's reformulation. Keep routine low-risk logging proportionate.
VII. Crowder v. LinkedIn: Court Compels Generative-AI Search Across Custodians
Case or Development Summary and Status
In Crowder v. LinkedIn Corporation, No. 4:22-cv-00237, a Northern District of California discovery order required LinkedIn to run a generative-AI search across the files of four custodians and produce responsive documents. The disputed string used terms including generative AI, GenAI, LLM, copilot, GPT and OpenAI near LinkedIn, Premium, member or data terms. The court found the request relevant to the plaintiffs' theories about data, machine learning and inference and found that LinkedIn had not supplied specific cost or time evidence sufficient to support its burden objection. The public summary of the order states that responsive production was due September 16, subject to extension if fact discovery moved. The court also permitted a limited reopening of one deposition on a separate issue. This is a discovery ruling about relevance and proportionality, not a finding that LinkedIn's AI conduct violated antitrust law.
Litigation AI GRC Analysis
For Litigation AI GRC, the order is a concrete example of 'AI' becoming an eDiscovery vocabulary rather than merely a subject for expert opinion. Search terms, custodian selection, burden evidence and preservation practices can determine whether AI-related records are captured. Enterprises that deploy multiple AI products may have relevant records across email, collaboration tools, source repositories, prompt logs, model configuration systems and vendor portals. Overcollection creates cost and privilege risk, but under-preservation can create larger litigation exposure. The control objective is a defensible, matter-specific information map.
Strategic Risk Management Recommendation
When AI becomes relevant to a dispute, legal teams should identify the systems and custodians actually connected to the pleaded theory before choosing search terms. Document proportionality decisions, test search strings on sample data, and record known gaps in prompt or platform logging. If a party objects on burden grounds, support the position with concrete volume, cost and timing evidence rather than generalized assertions. Preserve the distinction between search relevance and substantive liability. Where an AI deployment changes during litigation, update the data map and preservation scope rather than relying on the original collection plan.
VIII. xAI Seeks Appellate Relief Against Minnesota's AI 'Nudification' Law
Case or Development Summary and Status
Reuters reported that xAI asked the U.S. Court of Appeals for the Eighth Circuit to block enforcement of Minnesota's law addressing fake nude images while its constitutional appeal proceeds. The appeal is X.AI LLC v. Keith Ellison, No. 26-2806. Earlier in September, a federal district judge declined to grant preliminary injunctive relief, leaving the law in effect at that stage. xAI argues that the measure violates constitutional protections; Minnesota is defending the statute. The September 14 request is therefore an appellate interim-relief development, not a final determination that the law is constitutional or unconstitutional. The dispute sits at the intersection of synthetic media, non-consensual intimate imagery, platform design and speech rights, and it illustrates how states are testing traditional legal tools against new generative capabilities.
Litigation AI GRC Analysis
For Litigation AI GRC, the case shows that product controls can become evidence in constitutional and statutory litigation. Relevant records may include feature design, safety filters, user reporting, enforcement metrics, age or identity controls, incident responses and internal assessments of foreseeable misuse. At the same time, governance should not assume that a harmful-use concern automatically determines the constitutional validity of a particular law. Litigation teams need separate lanes for technical risk, statutory obligations, constitutional defenses and product-policy decisions. Interim injunction standards should also be kept distinct from final merits review.
Strategic Risk Management Recommendation
Platforms capable of synthetic image generation should maintain versioned records of safety controls, known abuse patterns, material policy changes and escalation pathways for non-consensual intimate content. Product and legal teams should map state-specific obligations rather than assuming federal uniformity. When challenging or defending regulation, preserve evidence supporting both the technical feasibility of safeguards and the practical burdens of compliance. Public communications should accurately distinguish a denied preliminary injunction, a pending appeal and a final constitutional ruling. Avoid describing a live challenge as settled precedent.
IX. Expert Witness Logs Highlight a New Discovery Record
Case or Development Summary and Status
Bracewell published a Law360 analysis of AI use by an expert in Montenegro Enterprises v. 3M Co., ongoing multidistrict litigation arising from a Houston gas explosion. The article emphasizes that prompt histories can expose the expert's process, objectives and iterative reasoning in ways that traditional drafting often did not. It also points to a separate May 2026 District of Connecticut order in Conservation Law Foundation v. Shell Oil Co. directing disclosure of AI prompts and queries used in preparing an expert report, notwithstanding a stipulation protecting ordinary expert notes, drafts and communications. The Bracewell piece is practice commentary. Its description of the 3M matter should not be restated as a new universal discovery rule.
Litigation AI GRC Analysis
For Litigation AI GRC, expert AI use raises three separate questions: reliability of the opinion, discoverability of the AI interaction, and the expert's ability to explain independent reasoning. Prompt logs can reveal leading instructions, target conclusions, omitted alternatives and reliance on facts that never entered the expert's own analysis. Deleting those records can also create preservation questions once discovery obligations attach. At the same time, not every AI-assisted task is necessarily discoverable; privilege, work-product, Rule 26 and party stipulations remain matter-specific.
Strategic Risk Management Recommendation
Expert engagement letters should address approved AI tools, confidential inputs, disclosure to retaining counsel, preservation of material prompts and outputs, and independent verification. Counsel should ask experts about AI use before the report is drafted, not at deposition. For opposing experts, discovery requests should be tailored to material AI use and governing procedural rules rather than assuming every interaction is discoverable. Preserve the expert's human reasoning, source materials and validation steps. If parties want special treatment for expert AI records, address it expressly in discovery stipulations rather than relying on generic draft protections.
X. Morgan & Morgan Commits $1 Billion to Legal AI and Its MX2 Platform
Case or Development Summary and Status
Reuters reported that Morgan & Morgan committed to invest at least $1 billion in AI and technology over the next decade and said it has already spent about $300 million building its proprietary MX2 platform. The firm says MX2 is used to extract medical information, generate case documents and assist trial preparation and has nearly 5,000 monthly users. Morgan & Morgan plans to make the platform available to other law firms on an invitation-only basis by the end of 2027. The report also notes that two Morgan & Morgan lawyers were sanctioned in a prior matter for fake cases and that the firm says it has since strengthened training and review processes. The current announcement is not proof that the platform eliminates hallucinations or satisfies every confidentiality, evidence or professional-responsibility requirement. It is a significant signal that litigation firms are moving from isolated AI subscriptions toward firm-controlled platforms, security investment and operational integration.
Litigation AI GRC Analysis
For Litigation AI GRC, the strategic issue is governance at scale. Once thousands of users rely on a platform for medical records, document generation or trial preparation, isolated reviewer habits are not enough. Firms need access controls, approved use cases, source traceability, testing, incident management, change control, training and metrics that show whether controls are working. Internal platforms can improve data control, but ownership does not by itself establish accuracy, privilege protection or evidentiary reliability. Vendor or firm claims should remain distinct from independent validation.
Strategic Risk Management Recommendation
Law firms building or buying litigation AI should define a risk-tiered control model before expanding user access. High-stakes functions should require authoritative-source verification, documented human review and an escalation path for material uncertainty. Procurement or internal development should address data retention, role-based access, audit logs, model changes, incident reporting and the ability to export records needed for disputes. Measure reviewer time, exception rates, hallucination incidents and override behavior rather than adoption alone. If a system is offered to other firms, clarify contractual responsibility and avoid implying that technical safeguards transfer professional judgment away from the user.
Author: Pouya Shafabakhsh Co-Founder, CAIO & Principal Forensic AI Auditor, Radsam Academy of AI Sovereign Governance. Independent AI governance and forensic-audit analysis. Educational information only; not legal advice.
Would you be opposed to finding your firm's AI gaps before opposing counsel or insurance companies do? I invite you to assess your firm using our AI Risks Calculator. Safely, securely, privately, and confidentially—with no need to enter any information, not even your email address—you can get a comprehensive AI Risks and Liabilities report in just five minutes.
If you have a sensitive file and want to protect it against AI hallucinations, bias, algorithmic discrimination, and AI-assisted risks through an Air-Gapped Shadow AI Audit before court, or if you are involved in a legal dispute related to AI and need a Judicial Forensic AI Audit, at Radsam, we are willing to assist you whether through our human-based services or by providing our in-house solution, the Sovereign Sanctuary Vault. In these cases, I invite you to fill out our Pre-Qualifying Assessment Form. We will review your information, qualifications, and capacity, and will contact you within two business days.
Related Articles
For the governing or strategic frame behind the september 15, 2026, Ontario AI Class Actions: Map Decisions Across Versions and Cohorts explains historical system state, cohort variation, and class-wide evidence.
To compare the closest operational issue raised by the september 15, 2026, AI-assisted eDiscovery controls examines record ownership, preservation, proportionality, and discovery scope.
For the evidence and reconstruction layer connected to the september 15, 2026, Prompt Injection Litigation: What Law Firms Must Prove focuses on the adjacent governance, evidence, and verification issue.
To connect the september 15, 2026 with governance or implementation choices, Ontario Arbitration: Set the AI Handling Boundary Before the Hearing covers AI record handling, confidentiality, and hearing readiness.
For the next practical decision after the september 15, 2026, New York Corporate Disputes: Separate AI Output from Human Decisions provides context on AI-assisted business decisions, diligence records, and dispute reconstruction.
Comments