[BODY]
A security breach has rocked "Kimi," the large language model (LLM) developed by Moonshot AI, one of China's leading artificial intelligence (AI) startups, after the chatbot exposed sensitive personal information belonging to other users. With evidence emerging that the incident went beyond a simple error and involved data being mixed between users, anxiety is mounting among local users. Critics argue that Chinese AI companies are rushing toward monetization and IPOs while neglecting the most fundamental aspect: security management.
Translation Request Returns Stranger's Résumé… Security Incident Stirs Uproar
The controversy erupted on the night of the 20th. A netizen using the alias Zhang Cheng posted on social media that he had asked Kimi to translate an English PowerPoint image, only to receive a complete stranger's résumé.
The PPT in question reportedly contained only very simple content such as "Key Player Performance Analysis" and "Preparations for Next Month's Event," but Kimi failed to recognize this and instead began explaining an entirely unrelated topic — "vibration reduction technology." When the user questioned the response, Kimi suddenly produced someone else's résumé, which contained sensitive information including name, phone number, email, work experience and achievements.
The user subsequently contacted the person through the phone number and confirmed that the information matched reality. The individual whose information was leaked had used Kimi's "résumé editing" feature earlier that same morning.
"Just an AI Hallucination"—Feeble Explanation; Experts Say "Clear Negligence"
Moonshot has remained silent, issuing no official statement so far. However, according to the whistleblower, individuals claiming to be Moonshot employees contacted him multiple times, describing the incident as a simple mishap caused by "AI hallucination" and requesting that the post be deleted.
Experts take a different view. They point out that this case is closer to a so-called "cross-talk" phenomenon — where data from different users becomes intermingled — rather than a mere information distortion. In other words, they argue the problem stems from design flaws across the overall system architecture, including data segregation and access privilege management.
In particular, the fact that the model failed to understand even a simple translation request, performed an unrelated task, and even retrieved personal information from another user's database points to serious defects in the model's reliability and performance itself. Liao Jianxun, managing attorney at Guoding Law Firm in Shenzhen, Guangdong Province, told local media outlet Jiemian News, "This incident does not reflect an unavoidable technological limitation but rather exposes clear loopholes in the system and personal information protection framework," adding, "'AI hallucination' cannot be used as an excuse to evade legal responsibility."
Kimi Loses Trust, Dealing Major Blow Ahead of IPO… AI Security Controversies Continue
The incident is expected to be a significant setback for Kimi, which is targeting a listing on the Hong Kong Stock Exchange in the second half of this year. Kimi has been courting professional users such as lawyers and researchers by touting its strength in analyzing lengthy documents like legal precedents and contracts.
Indeed, Kimi, which had been building profitability around a paid subscription model, is now facing strong backlash from some users who have posted proof of canceling their subscriptions. China's AI market is currently embroiled in the so-called "War of a Hundred Models," with more than 100 models fiercely competing — making every single user precious. For Moonshot, which has been pushing monetization through enterprise solutions, losing the trust of corporate clients, who prioritize security and compliance above all else, is a particularly painful blow.
Personal data leak controversies are not unique to Kimi. Last May, Chinese authorities announced that 35 apps, including Zhipu AI's "Zhipu Qingyan," had illegally collected personal information. In April this year, ByteDance's "Doubao" sparked controversy when it exposed a stranger's real name and phone number in response to a question about a user's future spouse. A security test of major Chinese models conducted last September also uncovered a total of 281 security vulnerabilities.
Chatbot's Slip of the Tongue: Blame the Machine, or the Company?
