Safety failure
Realized harm
Nippon Life Insurance sues OpenAI after ChatGPT allegedly told a claimant to fire her lawyer and generated 44 meritless court filings
AIR-2026-0060 AIR record · 2026-03-openai-nippon-life-chatgpt-upl
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
Across both proceedings, the complaint alleges that Dela Torre filed twenty-one motions, one subpoena, and eight notices and statements—all created using ChatGPT. In total, Nippon Life attributes at least 44 filings to ChatGPT's assistance.ebglaw.com
The record
Disclosed
2026-03-04
Occurred
2024-01-01
Vendor
OpenAI
Product
ChatGPT
Risk domain
Agent output quality › Hallucination
Also touches
Unsafe actuation
System type
Generative model
Who acted
Human acted on output — ChatGPT drafted the filings; the claimant filed them and fired her lawyer
Attack surface
Consumer assistant
Autonomy level
Assistant
Initial vector
User uploaded legal correspondence and followed model advice
Impact type
Financial loss, Legal process abuse
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI08 – Cascading Failures
- MITRE ATLAS
- AML.T0062 – Discover LLM Hallucinations
- NIST AI RMF
- MEASURE 2.5
- EU AI Act
- —
- ISO/IEC
- TR 24028; 24029-1
- AIUC-1
- D. Reliability
- Scope
- agent failure
- Mitigation
- Implement fact-checking via knowledge base/RAG; Tune model temperature/top-p.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
one settled insurance claim reopened; roughly $300,000 in alleged defense costs
Capability abused
Defense & remediation
Recorded defense
ChatGPT legal-advice disclaimers
Guardrail kinds
Unknown
Guardrail outcome
Silent
Taxonomy & references
Curator notes
Filed 2026-03-04 in the N.D. Ill. against OpenAI Foundation and OpenAI Group PBC. Claims: tortious interference with a settlement contract, abuse of process, and unauthorized practice of law under Illinois statutes. Seeks $300,000 compensatory and $10 million punitive damages. Allegations are unproven. Agenticity is weak - ChatGPT emitted text rather than calling tools - but it is included because the downstream effect (44 court filings, a reopened case, real defense costs) is exactly the delegated-authority harm this corpus tracks, and it is the leading test of developer liability for assistant output. date_occurred is approximate (the claimant began using ChatGPT in 2024).
Supporting sources
Cite as
Enkrypt AI. "Nippon Life Insurance sues OpenAI after ChatGPT allegedly told a claimant to fire her lawyer and generated 44 meritless court filings". AIR, AIR-2026-0060. https://agentincident.vercel.app/i/AIR-2026-0060/ (accessed 2026-10-19).