DeepSeek Accused of Involvement in Police Surveillance Project; Claude Becomes Data Relay Station

Overview

Recently, a tech community exposed that the artificial intelligence model DeepSeek is implicated in a controversy surrounding its alleged assistance with police surveillance projects. Meanwhile, Claude, Anthropic’s flagship model, is suspected of serving as a critical data relay hub. This report has sparked widespread public concern and ethical reflection on the application of large models in sensitive domains.

According to disclosures, the incident involves a well-known domestic large-model company, Moonshot AI (Kimi). Certain users with ties to the background of the Chinese People’s Liberation Army (PLA) exploited the platform to make requests for specific scenarios, subsequently redirecting data chains to other model services.

“AI models should not become backdoors for circumventing regulation and obtaining sensitive law-enforcement data.” — A cybersecurity researcher

Data Sources and Routing Paths

The report indicates that the core of the problem lies in the concealment of data routing. Some users assessed as having military or law-enforcement backgrounds initiated requests through the Moonshot AI platform, targeting real-time or delayed video streams from a large number of public surveillance cameras across Chengdu.

Specifically, these users did not directly perform complex video analysis on DeepSeek or Kimi backends; instead, they forwarded surveillance footage data and analysis requirements to Claude for processing. This “detour” approach aims to bypass the direct content-filtering mechanisms that domestic large models have in place for sensitive material.

Within this chain:

• DeepSeek is alleged to have been involved as part of engineering requests to construct the underlying logic for police surveillance systems.
• Moonshot AI (Kimi) served as the initial user-access point.
• Claude became the actual “black-box” node executing complex data parsing and relaying.

DeepSeek’s Engineering Connections

Beyond data relay, DeepSeek itself has also come under scrutiny. It is alleged that engineers engaged in architecting police surveillance systems submitted technical consultations or code-generation requests to DeepSeek regarding system construction.

Although DeepSeek has not yet issued a detailed official statement on this matter, such use of open-source or commercial large models to build law-enforcement surveillance tools easily ignites discussion regarding the neutrality of technology and the boundaries of misuse. An engineer need only generate a surveillance-algorithm framework through the model, then combine it with illegally obtained video sources, to assemble a covert surveillance network.

Ethical and Regulatory Warnings

This incident has once again sounded the alarm in the field of AI safety. As large-model capabilities advance rapidly, defining the boundary between public safety and privacy protection has become a challenge common to the global technology community.

This concern is especially pronounced when mainstream domestic models are used as stepping stones to route sensitive law-enforcement data toward overseas servers or third-party models, highlighting acute issues of data sovereignty and security vulnerabilities. Regulators urgently need to strengthen audits of API-call behaviors and explicitly prohibit the use of large-model services for unauthorized surveillance-data collection and analysis.