Alibaba, Claude Code decided to stop its use within the company and notify its employees July 10, 2026 He pointed out the date. Chinese technology giant uses Anthropic’s coding tool high risk software added it to the list. According to the news reported by Reuters, based on a source close to the subject, the company has hired the developers Qoder directs him to his vehicle. This decision will directly affect the daily working tools of coding teams.
Reuters reported that Alibaba attributed the decision to “backdoor risks” and security concerns. Similar justifications were included for Claude Code in the internal announcement seen by the South China Morning Post. However, Alibaba told its employees to abandon the outsourced tool and switch to the company’s own solution. Alibaba and Anthropic did not publicly respond to the information in the Reuters report.
Claude Code helps developers write code with natural language commands via terminal, IDE, or GitHub. The tool also handles tasks such as editing files, running commands, and managing Git workflows. Anthropic’s GitHub page describes Claude Code as a tool that understands the code base and handles routine development work. Therefore, Alibaba’s decision covers a broader area than a classic chatbot reach.
On the Anthropic front, the issue is directly linked to usage restrictions for Chinese companies. The company already bans Chinese companies and their overseas-owned subsidiaries from using models. In addition, Anthropic states that it has developed new measures to close the intermediary routes that provide access to Claude through China. The company says it does not offer commercial Claude access in China, explaining these restrictions with legal, regulatory and security risks.
Anthropic limits Claude access to China-based uses
The last leg of the discussion included technical claims shared on Reddit. According to these posts, a version of Claude Code included a mechanism that could secretly identify users with Chinese connections. Thariq Shihipar from Anthropic described this mechanism in his statement on X as an experiment that started in March. Shihipar is the team’s experiment to prevent account abuse through unauthorized sellers and model distillation He said that he conducted it to protect against these attempts.
Model distillation refers to the practice of training an AI model with the outputs of another model. In its article published in February, Anthropic accused some artificial intelligence laboratories in China of conducting model distillation attempts with Claude outputs. In addition, in the letter quoted by the Wall Street Journal, Anthropic claimed that Alibaba accessed Claude through unauthorized means and conducted a large-scale distillation operation. Alibaba did not respond to these allegations with a public statement.
Shihipar also stated in his statement that the team developed stronger measures for the experiment in question. Additionally, he said Anthropic has been planning to remove this mechanism for some time. The Register reported this statement in the context of the discussion of the hidden detection mechanism in the Claude Code. Thus, Alibaba’s security justification came under the same heading as Anthropic’s allegations of unauthorized access and distillation.
What Alibaba recommends to employees Qoderstands out as the company’s own development tool. According to Alibaba Cloud documentation, Qoder offers desktop IDE, CLI and JetBrains plugin support. The platform can be connected to Alibaba Cloud Model Studio via a pay-as-you-go model, Coding Plan or Token Plan. Thus, the company aims to keep developer workflows within its own ecosystem.
For Alibaba employees, the impact of the decision will be felt especially in the processes of writing code, debugging and editing projects. Teams that include Claude Code in their daily development work will turn to company-approved tools after July 10. On the other hand, the steps taken by Anthropic against Chinese-related accesses expand the security debate in artificial intelligence-supported coding tools. Coding agents are now evaluated not only on productivity, but also on company data and intellectual property.