American AI company Anthropic has been one of the most-vocal supporters of banning export of American AI software and hardware to China. So much so that it’s CEO Dario Amodei has called China an adversarial nation” on numerous occasions. Like Microsoft-backed OpenAI and Alphabet’s Google, Anthropic has not made its services available in mainland China, citing national security concerns. Due to this, China is among the list of countries, along with Russia, North Korea, Afghanistan, Iran and Cuba, where Claude chatbots and AI models are not available.Last week, a security researcher flagged that Anthropic accidentally made the source code for its popular vibe-coding tool, Claude Code, public. The news soon made developers start reposting the code on the developer platform GitHub. It triggered a frenzy among Chinese developers as well.The source code – running more than 512,000 lines and buried deep in the software package – was reportedly spotted and decrypted by software engineer and cybersecurity researcher Shou Chaofan, who then posted it on Twitter. According to a report in South China Post, Chinese developers have been having a field day since the leak.Several Developers in China are said to be scrambling to download copies of the leaked code and poring over the files to learn every detail. What reportedly makes Chinese developers highly enthusiastic about Anthropic’s AI models is their advanced coding capabilities. On Chinese forums, many shared what they deemed to be the secret recipe for Claude Code — from its architecture and agent design to memory mechanism, among others. One topic titled the “Claude Code source code leak incident” has more millions of views, with many local developers sharing what they had learned and suggesting how they could make better use of the tool.Though some industry experts claim that the leaked file only included codes for Claude Code, and not the model weights, there is also a view that says the leaked data is still a treasure trove for developers. As Zhang Ruiwang, a Beijing-based IT system architect, told South China Post, “But the code batches are indeed a treasure for AI companies or developers, as they revealed all the key engineering decisions Anthropic made.”
Anthropic accused Chinese companies of stealing Data from Claude
The leak comes just a month after Anthropic said that three Chinese AI companies set up more than 24,000 fraudulent accounts with its Claude AI model to help their own systems catch up. In February, in a blog post, Athropic claimed that three Chinese companies—DeepSeek, Moonshot AI and MiniMax—prompted Claude more than 16 million times, siphoning information from Anthropic’s system to train and improve their own products.
Poll
Should companies take stronger action against data theft in AI development?
Anthropic said distillation had legitimate uses—companies use it to build smaller versions of their own products, for example—but it could also be used to build competitive products “in a fraction of the time, and at a fraction of the cost.”
