August 2026
Watson Grinding v 3M
Expert asked ChatGPT to 'show how 3M is 0% at fault.' Plaintiffs obtained 350 pages of AI prompts. Jury awarded $61.5M. Nearly 2,000 additional plaintiffs in the pipeline.
404 Media, Forbes, AI Weekly
AI Risk Tracker
2,039
Court cases involving AI-fabricated content across 44 jurisdictions since 2023
180% growth in the last 12 months
805
Involving licensed lawyers
159
Professional sanctions issued
44
Countries and jurisdictions
$109,700
Largest single penalty to date
The governance gap
Adoption is only part of the story.
32%
of clients reconsidering firms without AI governance
59%
describe their approach to AI as cautious
Expert witness frontier
Sector focusWhat happens when AI enters expert evidence?
17%
of firms use AI to support expert witness work
7%
keep audit logs of that AI use
3%
check whether AI outputs would survive a court challenge
1%
have no formal AI controls at all
Secretariat and ACEDS, 2026. First year the survey included expert witness AI.
45%
plan to start using AI with experts within 12 months
12%
have a written policy for AI in expert work
10%
train their experts on AI tools
The wider risk landscape
Hidden instructions in documents can steer model output. Ecsper examines document integrity and prompt-injection exposure in your workflow.
The Ecsper Governance Index measures AI governance maturity across regulated professional services. It is built from anonymous assessment data contributed by professionals who take the Check Your Exposure assessment. No personal data is used.
The Governance Index is being built. Once enough professionals in a given sector have completed the assessment, anonymised benchmarks will appear here. Take the assessment to contribute.
Growth trajectory
How the numbers are changing.
~8
new court cases added every day
180%
Growth in the last 12 months
$109,700
largest single penalty to date
Cases over time
By jurisdiction
Lawyer vs pro se
Penalty distribution
'Not published' indicates cases where the court did not publish a monetary penalty. This does not mean no penalty was imposed.
By legal field
These are only the cases a court caught and wrote up. The true number is higher.
Explore the evidence
Case explorer
2,039 cases across 44 jurisdictions. Filterable, sortable, and searchable.
Tap a card to expand details
Page 1 of 1
The cases that set the precedent
Landmark cases
Five cases that have shaped how courts, regulators, and the profession respond to AI in professional work.
May 2026
CLF v Shell
Expert used GPT-4o to filter discovery documents. Court ruled the prompts are 'part of the expert's methodology' and discoverable under Rule 26(b). First federal court order requiring expert AI prompt disclosure.
Mayer Brown, Dechert
June 2026
Withers v City of Aberdeen
Both sides filed fabricated citations. Judge cancelled the trial and suspended both lead attorneys from the district for two years. New severity ceiling for AI sanctions.
HAQQ Legal AI
2025
Couvrette v Wisnovsky
Oregon winery dispute. Combined sanctions and fees reached approximately $109,700, believed to be the largest aggregate penalty for AI-fabricated filings.
ABA Journal
June 2023
Mata v Avianca
Lawyer asked ChatGPT for cases. Six did not exist. $5,000 sanction. Three years later, the same failure has its own public database and the database needs daily updates.
Court filings, Charlotin database
Provenance and interpretation
Sources and methodology
All figures in Ecsper AI Risk Intelligence are drawn from independently published sources. This page documents the provenance, scope, and update schedule for each.
Charlotin AI Hallucination Cases Database
The figures on this page are drawn from the AI Hallucination Cases database maintained by Damien Charlotin, a researcher at Sciences Po, Paris. The database is publicly available, licensed under CC BY 4.0 with attribution, and its inclusion bar is strict: only decisions where a court or tribunal explicitly found or clearly implied reliance on hallucinated material. Ecsper reviews and verifies the count quarterly against the live database. The number shown is rounded down to remain accurate between updates. Individual case details are verified against primary sources and reputable legal publications. This tracker counts documented outcomes. Cases where fabricated content was filed but not caught are, by definition, invisible. Whatever the true number is, this is the floor. The full dataset is available as a CSV download from the source database. Ecsper downloads, trims, and analyses the dataset monthly. The case explorer shows the trimmed dataset; for the complete record including hallucination item details and source documents, visit the original database.
Thomson Reuters Institute, 2026
Contextual statistics on AI adoption and governance gaps are drawn from the Thomson Reuters Institute Future of Professionals Report 2026, based on a survey of 1,816 professionals across 62 countries. These figures appear in the Overview dashboard. They are cited individually where they appear and are not drawn from the Charlotin database.
Secretariat and ACEDS, 2026
Expert witness AI adoption and controls data are drawn from the Secretariat and ACEDS 2026 Artificial Intelligence Report, a survey of legal professionals conducted in March 2026 across law firms, corporations, government agencies, service providers, and independent consultancies. This is the first year the survey included a dedicated section on AI use with expert witnesses.
Stanford RegLab, 2024
Error rates for commercial legal AI tools are drawn from a peer-reviewed Stanford RegLab study, which found hallucination rates of 17% to 34% even in paid platforms including Westlaw AI-Assisted Research and Lexis+ AI. This finding is cited in the growth trajectory section.
How to read the dashboard
Blue identifies adoption indicators; red identifies risk and control indicators. Percentages from separate surveys are not additive. Court cases, lawyers, and sanctions are different counting units.
Charlotin, D. AI Hallucination Cases Database. https://www.damiencharlotin.com/hallucinations/. Licensed under CC BY 4.0. Last verified: September 2026.
Thomson Reuters Institute. Future of Professionals Report 2026.
Dahl, M. et al. 'Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools.' Stanford RegLab, 2024.
Secretariat and ACEDS. 2026 Artificial Intelligence Report. July 2026.