Anthropic’s AI code leak ignites frenzy among Chinese developers

Anthropic has not made its services available in mainland China, citing national security concerns.

Photo: dpa Less than a year after US artificial intelligence start-up Anthropic called out China as an “adversarial nation” and vowed to restrict the country’s access to its technologies, the company inadvertently released coding secrets, triggering a frenzy among Chinese developers.

The leak occured after an Anthropic employee accidentally included the modified source code of Claude Code, the company’s sensational AI coding tool popular among developers worldwide, in a file within a software package hosted online.

The source code – running more than 512,000 lines and buried deep in the software package – was spotted and decrypted by software engineer and cybersecurity researcher Shou Chaofan on Tuesday, who then posted it on X.com.

By Thursday, it had more than 33 million views, drawing attention across the tech community globally and in China.

Like Microsoft-backed OpenAI and Alphabet’s Google, Anthropic has not made its services available in mainland China, citing national security concerns.

As a result, China is one of a handful of countries, along with Russia, North Korea, Afghanistan, Iran and Cuba, that cannot use Claude chatbots and AI models.

This illustration picture shows a computer screen displaying the Anthropic website in San Francisco, May 13, 2024.

Photo: dpa Despite the restrictions, Chinese developers appear to be highly enthusiastic about Anthropic’s AI models as well as Claude Code, thanks to their advanced coding capabilities that automate programming tasks, from writing code to debugging, while boosting development efficiency, even though they have to access the service via virtual private networks.

Chinese developers have been having a field day since Tuesday, scrambling to download copies of the leaked code and poring over the files to learn every detail.

Some of them shared what they deemed to be the secret recipe for Claude Code – from its architecture and agent design to memory mechanism, among others – in postings on social media platforms that subsequently generated a high number of views and triggered heated discussion.

One topic titled the “Claude Code source code leak incident” had more than 2.6 million views as of Thursday, with many local developers sharing what they had learned and suggesting how they could make better use of the tool.

Anthropic did not immediately respond to a request for comment on Thursday.

原文链接: 南华早报