Chinese Developer Turns Government Anti Fraud AI Into a Free OpenAI Compatible Service Before GitHub Removal

A Chinese developer briefly transformed a government backed anti fraud AI service into an OpenAI compatible interface that could be connected to popular third party AI applications, creating an unusual demonstration of how easily a specialized public service model can be repurposed when its underlying interfaces are exposed. The project, called fanzha-ai-proxy, appeared on GitHub under developer kltyton before the original repository disappeared only hours after attracting attention across Chinese developer communities.

The proxy was designed around National Anti Fraud AI, a mobile service focused on telecommunications and online fraud prevention. The official application is published by the Shanghai Institute of Criminal Science and Technology and provides AI assisted answers, fraud awareness information, and other anti scam resources. Its August 2026 version 2.0 update formally introduced the National Anti Fraud AI branding.

Rather than creating a new AI model, fanzha-ai-proxy acted as an intermediary between existing AI clients and the government service. Archived copies of the project show that it translated requests into the standard OpenAI chat completions format, making the underlying assistant compatible with software including NextChat, LobeChat, Codex CLI, and other applications already designed around OpenAI style APIs. It supported both normal and streaming responses while maintaining authenticated sessions through the service's existing token system.

The project documentation claimed authenticated sessions could remain usable for approximately 90 days through access and refresh token management. That capability effectively allowed the official anti fraud AI to behave like a general purpose backend for compatible applications instead of remaining accessible only through its intended mobile interface. The proxy code itself points toward infrastructure hosted under a Shanghai Public Security domain, reinforcing that this was not simply an unofficial recreation of the assistant but an interface communicating with the existing service.

The experiment did not remain online for long. Posts circulating in Chinese developer communities reported that the original GitHub project became unavailable roughly 5 hours after gaining wider attention. The original repository now returns a 404 response, although the developer's GitHub account remains accessible. There is currently no public evidence confirming why the repository disappeared, whether GitHub itself removed it, whether the developer deleted it, or whether any government authority requested action. Claims assigning a specific cause should therefore be treated as speculation.

Removing the original repository also did not eliminate the project. Copies and forks are still publicly visible on GitHub, with archived versions preserving the README and proxy implementation. That reflects one of the fundamental characteristics of open source distribution: once code has been copied and forked, removing the original repository rarely guarantees that every copy disappears with it.

The episode raises a broader security question surrounding government and enterprise AI services. Building a polished mobile application around an AI assistant does not necessarily mean the underlying service is isolated from other software. If authentication, rate limiting, API access controls, and client validation are not designed around the possibility of unofficial clients, developers may be able to reuse backend infrastructure in ways that were never intended by the service operator.

China is simultaneously increasing oversight of AI applications more broadly. The Cyberspace Administration of China said on September 2 that its latest enforcement campaign had removed more than 5.61 million pieces of illegal or harmful AI related information, taken action against approximately 49,000 accounts, and dealt with more than 2,400 websites and applications. The fanzha-ai-proxy situation is different from those content enforcement cases, but it illustrates another challenge regulators and service operators increasingly face as AI systems become accessible through interconnected APIs.

The most interesting part of fanzha-ai-proxy is not that someone found a way to access a free AI model. It is how little separation can exist between a purpose built government application and the reusable AI service underneath it.

OpenAI compatible APIs have become an industry standard of sorts because they make models extremely easy to connect with existing clients and developer tools. That interoperability is valuable, but the same flexibility can expose services to unintended use when authentication and access policies are designed primarily around the official application.

The original repository disappearing within hours shows how quickly attention can bring an experiment like this to an end. The surviving forks show the opposite side of the equation. Once source code reaches a public development platform, removing the first copy does not necessarily remove the technology.

Should government operated AI services use much stricter API controls, or is interoperability an acceptable tradeoff when the underlying service is already available to the public?