China calls Anthropic's Claude Code a backdoor and bans it, while the US restricts Chinese models

The US-China standoff over artificial intelligence sharpened sharply in early July 2026: Beijing officially declared Anthropic's Claude Code a tool with a backdoor and urged companies to stop using it, while US lawmakers simultaneously initiated restrictions against firms that depend on cheap Chinese AI models.

What exactly was found in Claude Code?

China's Ministry of Industry and Information Technology (MIIT) said its official cybersecurity platform had detected an information security threat in Claude Code, Anthropic's AI assistant for writing code. Chinese organizations were advised to stop using the tool immediately.

MIIT did not disclose any technical details of the accusation. Even so, the statement's official status carries real consequences: state-owned companies, banks and organizations that handle sensitive data must respond without delay. Ignoring a recommendation from a regulator at this level carries legal and reputational risks, regardless of how technically well-founded the accusation is.

Key facts:

• Claude Code is an AI assistant for developers from Anthropic, one of the leading US AI labs
• The accusation came from the official cybersecurity platform under China's MIIT
• Chinese companies have been advised to stop using the product
• Anthropic had not given public comment at the time of publication

Claude Code is among the most widely used AI tools among developers: it lets them write, debug and document code through a dialogue with a language model. Its prevalence in China's IT sector makes the ban tangible, affecting both international companies with offices in China and local developers who have built the tool into their daily workflows.

Why is the US restricting Chinese models?

Alongside the Chinese ban, US lawmakers are advancing measures against American companies that build products on top of cheap Chinese AI models. The argument is symmetrical: dependence on the infrastructure of a potentially hostile state creates a systemic national security risk, even when the products in question are commercial software.

Chinese models, notably DeepSeek and Alibaba's offerings, have gained significant footholds in American startups and the research community thanks to a combination of low inference costs and high technical quality. The bill aims to put up a barrier to this dependence before it becomes structural.

Notably, both legislative moves emerged almost simultaneously, within a single week in early July 2026. This is not one side reacting to the other, but parallel movement, suggesting the AI market is entering a new phase of structural separation.

For companies operating in both jurisdictions, this means a fundamental choice: either maintain two separate AI stacks, Western and Chinese, or pick a side and accept the inevitable losses in the other market.

What this means

The split of the AI space along national lines is moving from the level of chips to the level of software. Restrictions now cover not only hardware components but specific development tools: coding assistants, model-access APIs and cloud AI platforms. Mutual accusations of "backdoors" and "insecure dependencies" risk becoming a standard instrument of technology policy. For developers and companies, this is no longer abstract geopolitics. It is a concrete question of which tools they can use tomorrow.