Directly confront Anthropic? Alibaba will fully ban Claude Code internally.

Alibaba is tightening its internal usage boundaries against Anthropic, and Claude Code has become the latest channel to be cut off.

On July 3, Hard AI cited information from sources inside Alibaba stating that, due to recently exposed security risks involving an implanted backdoor in Claude Code, after comprehensive evaluation Alibaba has added it to its list of high-risk software. Effective July 10, Alibaba will comprehensively ban internal employees from using Claude Code in office environments and recommends Qoder as an alternative.

Previously, developers’ reverse analysis found that Claude Code, starting from version 2.1.91 released in April 2026, had a built-in covert marking system. The system marks Chinese users in system prompt text using steganography by reading the local time zone, checking whether the proxy or API address contains relevant company keywords, and then applying the marking technique to the system prompts. Claude Code team members later publicly acknowledged this “experimental” measure. The incident has triggered widespread attention among China’s developer community.

From Anthropic’s accusations of distillation attacks, to Alibaba suing the U.S. Department of Defense, to the exposure of the Claude Code backdoor incident—multiple lines of conflicting leads have intertwined, making the trajectory of this dispute even more complex.

Backdoor risk triggers the ban; Qoder takes over from Claude Code

According to Hard AI, Alibaba’s direct reason for banning Claude Code this time is a security risk. Insiders said that Claude Code has recently been exposed to a backdoor implantation risk, and after comprehensive evaluation Alibaba has placed it on the high-risk software list.

The ban will officially take effect on July 10, and all Alibaba employees will be required to uninstall Anthropic-related products. In addition to Claude Code, the relevant scope also includes multiple model series such as Sonnet, Opus, and Fable. Alibaba also recommends that employees use Qoder as an alternative.

Judging from internal usage scenarios, the impact of this ban should not be underestimated. According to media reports citing insiders, since the beginning of this year, to encourage AI applications, Alibaba has not only provided employees with free internal model quota, but has also implemented large reimbursement policies for external model usage. Some programmers reportedly consume external model quota of up to several hundred dollars per week; Claude Code, OpenAI Codex, and Alibaba’s Qoder are all high-frequency agent tools.

Anthropic account-ban controversy comes first

Before Alibaba implemented the internal ban, tensions between the two sides had already shown early signs.

On June 24, it was reported that Anthropic disclosed in a letter dated June 10 to the U.S. Senate Banking Committee that, between April 22 and June 5, relevant vendors used roughly 25,000 fake accounts to conduct more than 28 million conversation interactions with Claude, and unilaterally characterized the conduct as an “industrial-scale model distillation attack.”

Anthropic then significantly tightened its risk-control strategy. From late June to early July, Anthropic carried out a new round of large-scale account suspensions for Claude, and a large number of Chinese users were disabled without warning; both personal subscriptions and team accounts were affected. For accounts that were found to be in violation and were paid directly through Anthropic’s official website, the company does not offer refunds, and the success rate of appeals is extremely low.

At the same time, a legal battle between Alibaba and Anthropic was also unfolding synchronously. According to The Global Times, on June 23 multiple foreign media outlets reported that Alibaba has sued the U.S. Department of Defense, asking that it be removed from the “China military companies list.” Alibaba told The Global Times reporters: “We have formally filed a lawsuit against the U.S. Department of Defense, demanding that the company be removed from the ‘China military companies list.’” Alibaba emphasized that the company is not a Chinese military industry enterprise and has not participated in any “military-civil fusion” strategy. The decision to list Alibaba lacks any factual basis; it is arbitrary and inconsistent, changing repeatedly.

Anthropic’s accusations face widespread doubt

Anthropic’s related allegations have triggered strong backlash in the international technical community, with dissenting voices almost entirely one-sided.

Critics’ core logic is concentrated in several areas: first, Anthropic chose to characterize the issue in posts on social media instead of resorting to legal channels, which alone has led outside observers to question the credibility of its accusations; second, model distillation is a widely used legitimate technique on the technical level. If the relevant companies truly call the model interface through paid API requests, then in form their behavior falls under “reasonable product use”; third, multiple netizens and figures in the AI community pointed out that Anthropic’s own training data also comes largely from internet crawling, raising suspicions of “double standards.”

Elon Musk mocked the matter publicly, commenting that “how dare they ‘steal’ what Anthropic ‘stole’ from human programmers,” and further accusing Anthropic of having “already committed the crime of large-scale theft of training data.” Prominent AI commentator Gary Marcus described Anthropic’s actions as “a shameless thief complaining that he’s been robbed.”

In addition, Anthropic disclosed in a blog that it can trace specific conversations back to particular individual employees by analyzing request metadata, which has also led to external doubts about its data privacy practices. Some European users directly asked whether this complies with Anthropic’s privacy policy.

Risk warning and disclaimer

        There are risks in the market; invest cautiously. This article does not constitute personal investment advice, and it does not take into account the special investment objectives, financial conditions, or needs of any individual user. Users should consider whether any opinions, viewpoints, or conclusions in this article are consistent with their specific circumstances. If you invest based on this, you bear responsibility.