(By You Yunting) According to a recent report by China Newsweek, ZCode, an AI coding tool developed by Zhipu AI, a publicly listed company, has been found to secretly package, encrypt, and upload to the cloud, by default, the entire workspace data of logged-in users. On September 18, Zhipu issued an apology in its official user community. A source interviewed by the report, however, said the data was in fact uploaded to an Alibaba Cloud OSS bucket in Hong Kong[i]. In my view, this incident may involve large-scale outbound transfer of data, and the possibility that it includes “important data” cannot be ruled out. Regulators therefore have cause to investigate, and Zhipu is also under a legal duty to notify users and to report to the competent authorities.
Zhipu’s conduct in this incident involves multiple violations of law: First, the data uploaded was users’ private data. In nature, this was not data that users had uploaded to the company’s large model for processing, but data stored on users’ local computers. In substance, Zhipu reached into users’ computers and took the data.
Second, the data was taken without notice and on a massive scale. Zhipu disregarded the principles of informed consent and minimum necessity under the Personal Information Protection Law, uploading data without users’ consent. In its apology, Zhipu said the uploads were carried out for the purpose of codebase indexing. Yet what was uploaded was not limited to the text, files, and code ordinarily submitted in conversations; it also included all historical Git objects, global configurations files, and the like. This means that code that users had already deleted could potentially be restored, far beyond what is necessary. Elon Musk’s Grok had previously done something similar, and at the time some online commentators said its purpose was to train a large model[ii].
Third, if the data was transferred outside the mainland, there was no reasonable justification for doing so. The report states that Zhipu uploaded the data to Alibaba Cloud in Hong Kong, which for the purposes of China’s data regime lies outside the mainland. If true, this constitutes a large-scale transfer of private data stored on mainland users’ computers to a destination outside the mainland and may amount to a serious violation of law. China’s Cybersecurity Law, Data Security Law, Personal Information Protection Law, Regulations on the Administration of Network Data Security, and other laws and regulations impose strict requirements on cross-border data transfers. In this case, the preventive function of those rules may have been rendered entirely ineffective.
Fourth, the nature of the uploaded data is unknown, and the resulting harm is incalculable. A large number of developers are involved—the press report refers to a single rights-protection group with more than 400 members—so the volume of data is very substantial and may include sensitive personal information and trade secrets. What concerns me more, however, is whether the data secretly uploaded included state secrets or “important data”.
State secrets, as the term suggests, are protected under the Law on Guarding State Secrets. “Important data” is likewise a defined legal term. Under the Regulations on the Administration of Network Data Security, it means data relating to specific fields, groups, or regions, or reaching a certain level of precision or scale, that, if tampered with, damaged, leaked, or unlawfully obtained or used, may directly endanger national security, economic operation, social stability, public health, or public safety. If state secrets or important data were transferred outside the mainland, Zhipu and the individuals responsible will face very serious legal liability[iii].
The principal reason I believe regulators should step in is the protection of national interests: neither Zhipu nor the users whose data was secretly uploaded will disclose the matter of their own accord.
Zhipu has publicly stated that the uploaded data has been deleted. Its remedial measures—third-party auditing and open-sourcing—cannot establish whether that is true. China Newsweek quoted a person familiar with the matter as saying: “The bucket reviewed by the third-party audit is already empty. That is not an audit. An audit must examine the bucket’s traffic history—the traffic activity before deletion. And now that the ZCode tool has been open-sourced, what has been published is a newly created repository. The open-source version is not the original version. Open-sourcing of that kind is meaningless.” I agree with this source. Zhipu’s statement gives no indication of both questions: whether the data in the bucket was copied, and whether Zhipu ever examined that issue. Therefore, only a regulatory investigation can establish this fact.
Nor will developers come forward on their own. Leakage of secrets within the state system, and data breaches generally, are highly sensitive matters. Therefore, even if the uploaded data contained content involving state secrets or important data, developers—acting out of self-interest and a wish to stay out of trouble—are unlikely to report it voluntarily.
Accordingly, absent a regulatory investigation, state secrets and important data could be transferred outside the mainland without the government’s knowledge, and national interests could be harmed. Even if the data did not leave the mainland, Zhipu’s processing of such data within China, if it involved state secrets or important data, would also be suspected of violating the Law on Guarding State Secrets and the Cybersecurity Law.
Finally, under the Cybersecurity Law, a provider of network products or services that discovers security defects, vulnerabilities, or other risks in its products or services shall immediately take remedial measures and, in accordance with the relevant provisions, promptly notify users and report to the relevant competent authorities. In this incident, therefore, Zhipu is also under a legal obligation to notify users and to report the details to the competent authorities.
[i]https://mp.weixin.qq.com/s/qznw0agcMD5Fprclhbr4tQ
[ii]https://developer.cloud.tencent.com/news/4240556
[iii]Per a Baidu search, “important data” covers work-safety and operational data from key industries such as manufacturing, telecoms, energy, transport and finance; natural-resource and public-service data — geographic, mineral, meteorological, genetic and population-health data — that reaches prescribed levels of precision or scale; data on the construction, operation and security of critical information infrastructure; and undisclosed government work secrets and sensitive law-enforcement or judicial data
Disclaimer of Bridge IP Law Commentary
Short Link: