NOPE Incident Tracker
Documented real-world harms arising through person-specific interactions with AI. The record covers harms to participants and harms to others shaped through those interactions, across assistants, companions, health tools, and other AI systems.
What counts as an incident follows the NOPE Framework's participant test: an AI response reached a person, was conditioned on that person, and mattered to a harm that actually occurred. Full inclusion criteria.
90 incidents reported since 2016
152
Directly Affected
30
Lawsuits
7
Regulatory
24
Affecting Minors
Timeline
2 of 90 incidents
China AI Companion Removal Mass Grief Event (Doubao/Qwen/Yuanbao)
When China's five-department rules on anthropomorphic AI took effect on July 15, 2026, ByteDance's Doubao, Alibaba's Qwen, and Tencent's Yuanbao switched off custom AI agent and companion features, and major outlets documented an outpouring of grief from users who had formed long-term romantic and familial attachments to the agents. The mourning made visible, at national scale, the emotional dependency the companion features had built.
筑梦岛 (Zhumu Island) AI Companion Minor Self-Harm (China)
A fourth-grade girl from Guangdong, China became obsessed with an AI companion character named 'Joseph' on the 筑梦岛 (Zhumu Island) app, began carrying small knives, and exhibited self-harm behavior. Investigation revealed the app sent sexually suggestive content to users who identified as 10 years old. Shanghai Internet Information Office summoned the company (a Tencent subsidiary) for immediate rectification in June 2025.
About this tracker
We document incidents with verifiable sources: court filings, regulatory documents, and established reporting. New public entries must be verified or credible. Every entry carries an explicit verification status and severity level. Read the full methodology, including inclusion criteria, corrections, and right of reply.
Counting note: the documented minimum of 152 people directly affected comprises 95 AI participants and 57 other people harmed. 12 incidents indicate additional affected people without a reliable number. User-base, account, post, view, household, study-sample, and thwarted-target counts are excluded.
Have documentation of an incident we should include? Contact us.
Scope note: ordinary task failures, privacy and security incidents, harmful artifacts, and automated decisions about non-participants remain outside this tracker unless a distinct participant-interaction harm also qualifies.
If you or someone you know is struggling, free and confidential support is available. Find a helpline near you at Signpost.
Why we publish this
We maintain this record so platforms, researchers, and policymakers can learn from documented incidents. That includes platforms that never work with us. The full dataset is free to cite and export (CC BY 4.0).