AI INCIDENT ARCHIVE / CASE 1728
← Portfolio
AI INCIDENT ARCHIVE CASE 1728
CASE 1728
DOCUMENTED INCIDENT LEGAL / GENERATIVE AI
ARCHIVED CASE 1728

AI INCIDENT REPORT

The Citation
That Didn't Exist.

A case study examining how unchecked AI-generated information entered a professional legal filing.

CASE 1728
CATEGORY GENERATIVE AI
SECTOR LEGAL
STATUS DOCUMENTED
FILE / 1728 SCROLL OR OPEN TO INVESTIGATE

CASE FILE / 1728

Incident Archive

Select a document to examine the case.

CASE 1728 CASE SUMMARY / DOCUMENT 01

01 / INCIDENT

What
happened?

INCIDENT TYPE UNVERIFIED
AI OUTPUT
CASE SUMMARY

A court filing prepared with assistance from generative AI contained citation and quotation errors that were not adequately verified before the filing was submitted.

CASE Hill v. Foundation Media LLC
COURT U.S. District Court
Southern District of New York
AI TOOL Harvey
LAW FIRM Shapiro Arato Bach LLP
JUDGE Arun Subramanian
OUTCOME No sanctions imposed
INCIDENT RECORD 01.01

The incident arose in Hill v. Foundation Media LLC, a copyright case in federal court in New York. Attorneys representing Foundation Media submitted a filing that contained problems with legal citations and quotations. The court questioned whether unchecked use of artificial intelligence had contributed to those errors (Scarcella, 2026).

Cynthia Arato, an attorney representing Foundation Media, later told the court that another lawyer at Shapiro Arato Bach had prepared the initial draft and finalized the filing. According to Arato, the lawyer failed to properly verify AI-generated material before submission. The firm subsequently disclosed that tools developed by Harvey had been used to assist in preparing the filing (Scarcella, 2026).

Judge Arun Subramanian ultimately declined to sanction the attorneys, but used the incident to warn about relying on AI-generated legal work without careful human review. His September 29, 2026 order emphasized that citations, quotations, and legal propositions produced with AI assistance must be checked for accuracy (Hill v. Foundation Media LLC, 2026).

CASE NOTE / 1728 AI assistance does not eliminate human verification. Assessment explored in Document 05.
AI INCIDENT ARCHIVE 1728 / 01
CASE 1728 EVIDENCE LOG / DOCUMENT 02

02 / EVIDENCE

Examine the
record.

LOGGED 03
EXHIBITS
VERIFICATION RECORD

The incident can be reconstructed through independent reporting and court records. Together, these sources establish what was filed, how AI was used, how the attorneys explained the errors, and how the court responded (Scarcella, 2026; Hill v. Foundation Media LLC, 2026).

EXHIBIT A
NEWS REPORT OCT. 2, 2026

INDEPENDENT REPORTING

Reuters

Reuters reported on Judge Arun Subramanian's response to the AI-related errors and documented the attorneys' explanation of how the filing had been prepared.

WHAT IT ESTABLISHES
  • The filing contained citation and quotation errors.
  • The court questioned whether unchecked AI use contributed to the errors.
  • Cynthia Arato said another lawyer at her firm prepared and finalized the filing.
  • The firm later disclosed that Harvey's AI tools had been used in preparing the filing.
SOURCE 01 SECONDARY SOURCE
EXHIBIT B
ATTORNEY DECLARATION COURT RECORD

PRIMARY SOURCE

Declaration of Cynthia Arato

The declaration provides the firm's explanation of how the errors occurred and describes its response after the problem was identified.

DOCUMENT PREVIEW ATTORNEY DECLARATION

Declaration of Cynthia S. Arato submitted in Hill v. Foundation Media LLC. Scroll inside the document to examine the complete filing.

VIEW FULL DOCUMENT ↗
WHAT IT ESTABLISHES
  • Arato attributed the errors to another lawyer involved in drafting and finalizing the filing.
  • She stated that AI-generated citations had not been properly checked before submission.
  • She described the conduct as contrary to the firm's policies and training.
  • The firm said it was implementing additional safeguards following the incident.
SOURCE 02 PRIMARY SOURCE
EXHIBIT C
COURT ORDER SEPT. 29, 2026

PRIMARY SOURCE

Order of Judge Arun Subramanian

The court's order records its response to the incident and explains the broader concerns raised by AI-assisted legal work.

DOCUMENT PREVIEW FILING 57 / PAGE 1
Page one of Judge Arun Subramanian's September 29, 2026 order in Hill v. Foundation Media LLC VIEW FULL DOCUMENT ↗

Filing 57, page 1. Judge Arun Subramanian's order addressing the use of AI in the Foundation Media filing.

WHAT IT ESTABLISHES
  • The judge declined to impose sanctions.
  • The incident was treated as a warning about unchecked reliance on AI-generated legal work.
  • Lawyers were reminded that AI-assisted citations, quotations, and legal propositions must be checked for accuracy.
  • The order also raised concerns about the effect increased AI reliance could have on the training of junior lawyers.
SOURCE 03 PRIMARY SOURCE
✓
VERIFICATION STATUS Incident corroborated across reporting and court records.

Claims from individual participants are identified as their statements rather than treated as independent findings of fact.

AI INCIDENT ARCHIVE 1728 / 02
CASE 1728 INCIDENT RECONSTRUCTION / DOCUMENT 03

03 / WHAT WENT WRONG

Reconstruct the
failure.

ANALYSIS PROCESS
FAILURE
RECONSTRUCTION

The incident was not the result of a single action. The available records show a sequence in which AI-assisted legal work moved through the drafting process without adequate verification before reaching the court (Scarcella, 2026).

CASE 1728 / FAILURE POINT

WHAT WENT
WRONG?

FAILURE POINT HUMAN
VERIFICATION

AI-assisted material passed through the professional workflow without being adequately checked before submission.

INCIDENT RECONSTRUCTION CASE 1728
03.01
RECONSTRUCTION FINDING The critical checkpoint was verification.

The available evidence points to a breakdown between AI-assisted drafting and final human review. The technology produced or assisted with material that required verification, but the professional workflow failed to catch the problems before submission (Scarcella, 2026).

AI INCIDENT ARCHIVE 1728 / 03
CASE 1728 CASE DISPOSITION / DOCUMENT 04

04 / AFTERMATH

What happened
next?

CASE STATUS NO
SANCTIONS
AFTER DISCOVERY

Once the citation and quotation problems were brought before the court, the incident shifted from a filing error into a broader examination of professional responsibility, AI safeguards, and the role of human review.

01
COURT REVIEW SEPT. 2026

The errors were questioned.

The court identified problems involving citations and quotations in Foundation Media's filing and questioned whether unchecked use of artificial intelligence had contributed to them (Scarcella, 2026).

COURT INTERVENTION
02
FIRM RESPONSE COURT RECORD

The firm responded.

Cynthia Arato apologized to the court and explained that another lawyer at Shapiro Arato Bach had worked on the initial draft and finalization of the filing. Her declaration stated that the AI-generated citations had not been properly verified (Scarcella, 2026).

EXPLANATION SUBMITTED
03
COURT DECISION SEPT. 29, 2026

The judge declined sanctions.

Judge Arun Subramanian decided that no further action would be taken against the attorneys. However, he described the incident as a "wakeup call" and emphasized that AI-assisted legal work must still be carefully checked by lawyers (Hill v. Foundation Media LLC, 2026).

NO FURTHER ACTION
04
RESPONSE & SAFEGUARDS AFTER INCIDENT

New safeguards were promised.

Arato stated that the conduct was contrary to the firm's existing policies and training and that additional safeguards would be implemented following the incident (Scarcella, 2026).

PROCESS REVIEW
04.01
THE WARNING WENT BEYOND THIS CASE

AI accuracy was only part of the concern.

Judge Subramanian also raised a longer-term concern: increasing reliance on AI could interfere with the development of junior lawyers if they no longer gain experience researching, reasoning through, and drafting legal arguments themselves (Hill v. Foundation Media LLC, 2026).

CASE DISPOSITION INCIDENT 1728
SANCTIONS NOT IMPOSED
PROFESSIONAL WARNING ISSUED
WAKEUP
CALL
AI INCIDENT ARCHIVE 1728 / 04
CASE 1728 ANALYST MEMO / DOCUMENT 05

05 / MY ASSESSMENT

The human
checkpoint.

DOCUMENT TYPE PERSONAL
ANALYSIS
ANALYST NOTE

The following section represents my personal assessment of the incident based on the evidence presented throughout this case file.

05.01 / PRIMARY FINDING
The failure was not using AI. The failure was trusting its output without verification.
MY ANALYSIS CASE
1728 PERSONAL ASSESSMENT

In terms of the problem revolving around this case, it wasn't the use of AI but the misuse of it by taking it at it's word, even though the Harvey AI platform is designed specifically for use within the legal system it doesn't mean it's perfect and this isn't the only instance of this happening (RAGbase Legal Research Team, 2026). There are multiple examples like Mata v. Avianca, Inc where AI made up cases that were used to file legal motions (Goswami, 2023). and the only way to fix this is through the use of human verification. Verification is one of the core tenants of proper AI usage and it could've avoided many situations like these if someone took the time to verify the validity of the information provided by AI.

05.02 / THE CHECKPOINT

AI output should be a starting point.

01 GENERATE Use AI to assist with research or drafting.
→
02 QUESTION Treat the output as information to evaluate.
→
03 VERIFY Check claims against reliable sources.
→
04 USE Only rely on information after verification.
05.03
CONNECTION / PROMPTING PLAYBOOK

Verification is part of using AI well.

This incident directly connects to the verification principle in my AI Prompting Playbook. Regardless of how specialized or capable an AI system may be, its output still needs to be checked before it is relied upon.

VIEW PROMPTING PLAYBOOK ↗
FINAL ASSESSMENT CASE 1728

AI can assist the work. It cannot replace responsibility for the work.

ANALYSIS COMPLETE 05 / 1728
PERSONAL ANALYSIS 1728 / 05
CASE 1728 REFERENCE INDEX / DOCUMENT 06

06 / SOURCES

Follow the
evidence.

REFERENCE FILE SOURCE
INDEX
SOURCE RECORD

The following sources were used to document, verify, and analyze the events described in this incident report. Primary court records, independent reporting, industry analysis, and related incidents were used to examine the case from multiple perspectives.

01
INCIDENT DATABASE CASE RECORD

AI INCIDENT DATABASE

Incident 1728

INCIDENT

Used to identify Incident 1728 and establish the AI incident record examined throughout this report.

APA REFERENCE

AI Incident Database. (2026). Incident 1728: Law Firm Shapiro Arato Bach reportedly filed fictitious case citation and misattributed quotations using Harvey's legal AI platform in New York federal court.

OPEN INCIDENT RECORD ↗
02
SECONDARY SOURCE NEWS REPORTING

REUTERS

Judge warns AI could stunt lawyers' training and harm their clients

REPORTING

Used for independent reporting on the incident, the attorneys' response, Harvey's involvement, the court's decision, and the broader concerns raised about the use of AI in legal work.

APA REFERENCE

Scarcella, M. (2026, October 2). Judge warns AI could stunt lawyers' training and harm their clients. Reuters.

OPEN SOURCE ↗
03
PRIMARY SOURCE COURT ORDER

U.S. DISTRICT COURT

Hill v. Foundation Media LLC — Filing 57

PRIMARY

Used to document Judge Arun Subramanian's response to the incident, his decision to take no further action, his warning about verifying AI-assisted legal work, and his concerns about AI's effect on the training of junior lawyers.

APA REFERENCE

Hill v. Foundation Media LLC , No. 25-CV-5947 (AS), Dkt. 57 (S.D.N.Y. Sept. 29, 2026).

OPEN COURT RECORD ↗
04
SECONDARY SOURCE LEGAL REPORTING

COURT CAST

AI May Save Law Firms Time, But Judge Warns It Could Cost Young Lawyers Valuable Training

REPORTING

Used as additional reporting on the court's concerns about AI reliance, professional responsibility, and the effect AI could have on the development and training of younger attorneys.

APA REFERENCE

Court Cast. (2026). AI may save law firms time, but judge warns it could cost young lawyers valuable training.

OPEN SOURCE ↗
05
SECONDARY SOURCE INDUSTRY ANALYSIS

RAGBASE

Harvey AI Problems: What Law Firms Are Actually Complaining About

ANALYSIS

Used to provide additional context on the limitations of specialized legal AI systems and the continuing importance of human verification when using AI-generated legal information.

APA REFERENCE

RAGbase Legal Research Team. (2026, August 2). Harvey AI problems: What law firms are actually complaining about. RAGbase.

OPEN SOURCE ↗
06
SECONDARY SOURCE RELATED INCIDENT

CNBC

ChatGPT cited 'bogus' cases for a New York federal court filing. The attorneys involved may face sanctions.

RELATED

Used to support the Mata v. Avianca example discussed in my personal assessment as another instance in which unverified AI-generated legal information entered a federal court filing.

APA REFERENCE

Goswami, R. (2023, May 30). ChatGPT cited 'bogus' cases for a New York federal court filing. The attorneys involved may face sanctions. CNBC.

OPEN SOURCE ↗

PRIMARY Original court records and official case documents.

SECONDARY Independent reporting and analysis used for context and corroboration.

RELATED Additional AI incident used to support the personal assessment.

END OF CASE FILE CASE 1728

Evidence reviewed. Sources documented. Analysis complete.

REFERENCE FILE 1728 / 06