Alibaba Imposes Full Ban on Claude Code Use from July 10 Amid Backdoor Concerns

Alibaba Group announced it will completely prohibit employees from using Anthropic's AI tools starting July 10. This measure follows recent concerns over security and personal information leakage after hidden tracking code was discovered in Claude code, which was attempting to identify Chinese users. Alibaba cited the hidden tracking code found in Claude code as posing a backdoor risk for the ban. An insider at Alibaba confirmed to Chinese media outlet Caixin on July 3 that Claude code had been added to the company's list of high-risk software. Alibaba considers such AI tools 'high-risk software' and has instructed employees to delete Anthropic's models and agent products and switch to its in-house developed AI assistant, 'Qoder'.

An engineer from Anthropic acknowledged the existence of the code on social media and stated it would be removed from the distribution the following day. Anthropic explained that the hidden code included in Claude code was an experiment to prevent account misuse and data extraction. Anthropic clarified it was a measure to prevent fraudulent use of their accounts.

In a letter to US senators, Anthropic claimed that Alibaba-affiliated organizations operated approximately 25,000 fraudulent accounts, interacting with Claude over 28 million times. Anthropic asserted that operators linked to Alibaba's large language model research organization ran about 25,000 fake accounts. These accounts generated over 28.8 million interactions with Claude, with the apparent goal of mass-extracting Claude's response data to accelerate the learning of competing models.

Anthropic is restricting sales to companies controlled by the Chinese government and is identifying and blocking accounts that violate these terms. Anthropic is blocking bypass routes to prevent unauthorized access to Claude by Chinese companies such as Ant Group and ByteDance.