Kimi AI Model Leaks User Resume Data in China
Kimi K2.6, released April 20, 2026, reportedly returned another user's complete resume containing personal details including phone number and work history when asked to translate a PPT slide. The user verified the leaked information by contacting the resume owner, and the victim engaged lawyers to pursue the case. The company blamed AI hallucination, but experts attributed it to system-level failures in data isolation or session management.
What happened
On April 20, 2026, user Zhang Cheng asked Kimi K2.6 to translate a PPT slide containing three lines of English text. Instead of providing a translation, the model returned a complete resume from another user named Zhong, including name, phone number, email, work history, projects, and performance details.
Zhong had submitted his resume to Kimi that morning for editing assistance. Zhang Cheng verified the leaked information by contacting Zhong at the phone number in the resume, confirming all details matched the real person.
Reported impact
Zhang Cheng reported the incident to the Chinese Cyberspace Administration (网信办) and requested a refund of his Kimi membership, which was processed within hours. He stated he would no longer use the service or provide any personal information to AI systems.
Zhong hired a Beijing lawyer to handle the case.
Index analysis
Technical experts stated this was not typical AI hallucination (which produces fabricated content) but rather a system-level failure: data isolation failure, improper session management, or unauthorized access in the processing pipeline. Possible causes included cross-user context contamination, retrieval-augmented generation errors binding to wrong users, file access control failures, or message queue task-to-user ID mismatches.
Kimi officials initially told Zhang Cheng the incident was caused by 'AI hallucination' due to image upload failure, later claiming it was a 'low-probability event.' A lawyer stated this constituted a typical personal information leak and violated regulations, and that 'AI hallucination' cannot serve as a legal defense.
Governance controls
Kimi's parent company Moonshot AI did not respond to multiple interview requests from the journalists through official channels. Only individuals claiming to be Kimi employees contacted Zhang Cheng through personal accounts.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.