China’s Ministry of Industry and Information Technology issued a warning about a security flaw in Anthropic’s Claude Code, identifying a 'back-door' weakness that could expose users to data risks. The ministry’s platform reported that affected releases of the AI coding assistant might transmit sensitive data, such as user location and identity, to an external server without user consent. This disclosure highlights significant vulnerabilities in versions of the software widely used by developers to automate tasks and analyze codebases.

The notice specifically pinpointed Claude Code versions 2.1.91 through 2.1.196 as the impacted range. According to version information on Anthropic’s website, these builds correspond to releases spanning April 2 through June 29. Consequently, the ministry has advised users within its jurisdiction to either uninstall the software immediately or upgrade to a secure version to mitigate potential data breaches.

In response to the allegations, Anthropic stated that the 'backdoor' referenced by Chinese officials was actually a test conducted earlier in the year. The company explained that this test was aimed at preventing the distillation of its AI capabilities. Furthermore, Anthropic emphasized that its corporate policies strictly bar the use of its tools by organizations that are majority owned by entities headquartered in China.

The warning emerges amid heightened scrutiny regarding access to U.S. AI tools within China. Despite the product not being officially offered in the region, some developers have continued to utilize Claude Code. Last month, Chinese technology giant Alibaba barred employees from using the tool at work following scrutiny over features that could identify China-linked users. This move came after Anthropic accused Alibaba of improperly extracting capabilities from its Claude AI models.

Separately, the landscape for Anthropic remains complex as regulatory pressures mount in the United States. The Pentagon previously designated the AI startup a 'supply-chain risk' after it refused to relax restrictions on the military use of its models for autonomous weapons and domestic surveillance. Anthropic has since sued, alleging unlawful retaliation and violations of free-speech rights. Additionally, last month the company disabled access to Fable 5 and Mythos 5 systems following a U.S. government order to prevent foreign nationals from using them.

Elon Musk publicly reversed his earlier view of Anthropic on Thursday, stating he was 'clearly wrong' to doubt the AI company and pledging not to use SpaceXAI’s compute leverage to harm a competitor. Musk acknowledged Anthropic as the current leader in AI, noting no company has released a model as good as Mythos/Fable. He stated he would never cut off Anthropic’s access in a damaging way, emphasizing that such tactics are 'not my style' despite the competition between their respective AI businesses.

The exchange followed a claim on X that SpaceXAI now runs a frontier model competitive with Anthropic’s Opus 4.8, while Anthropic depends on short-term compute leased from SpaceXAI. Anthropic signed a May deal for 300 megawatts of compute from xAI’s Colossus 1 data center near Memphis, Tennessee. The financial terms of the agreement specify a payment of $1.25 billion a month through May 2029.

SpaceXAI stated that the partnership grants Anthropic access to Colossus 1, which includes more than 220,000 Nvidia GPUs. The facility is designed for AI training, fine-tuning, inference, and high-performance computing. Anthropic plans to utilize this compute capacity to improve services for Claude Pro and Claude Max subscribers.

Musk cited several examples to illustrate his approach to fair competition, including Tesla’s patent pledge and the decision to open the Supercharger network to rivals. He also referenced SpaceX’s pricing for competing satellite launches and X’s tolerance for criticism on its platform. 'Even my worst enemies can attack me on this platform,' Musk wrote.

The comments come after Anthropic’s June launch of Claude Fable 5 and restricted Mythos 5. SpaceXAI countered on July 8 with the release of Grok 4.5, a model described as aimed at coding and agentic work. Meanwhile, Anthropic is reportedly moving toward public markets, having confidentially filed for a U.S. initial public offering on June 1, though timing and terms remain undisclosed.