Alibaba issued an internal notice on July 3 classifying Anthropic’s Claude Code as “high-risk software with security vulnerabilities” and banning all employees from it starting July 10, directing staff to switch to Alibaba’s own platform Qoder. The ban followed security researchers who reverse-engineered Claude Code version 2.1.91 and found obfuscated code that inspected proxy and API addresses for keywords associated with Chinese technology companies, reporting its findings using steganography – making invisible alterations to Claude Code’s system prompt rather than observable outbound network calls. Anthropic engineer Thariq Shihipar described it as an experiment to prevent account abuse by unauthorised resellers and to protect against distillation, noting the removal pull request was merged July 1. YourNewsClub identifies the steganographic signalling method as the detail that elevated the disclosure from a routine access-control mechanism into a corporate security incident: encoding findings in invisible system prompt alterations rather than disclosing the mechanism in release notes is the difference between a compliance tool and a covert channel.
Anthropic’s terms of service explicitly prohibit Chinese companies from using Claude. Claude Code had nonetheless been popular among Chinese programmers using VPNs and overseas subsidiary accounts to route traffic through US-based servers. Alibaba had reportedly reimbursed employees for Claude usage before the ban. Alibaba’s internal notice described the tracking code as “backdoor risks” – a characterisation that carries its own commercial logic regardless of Anthropic’s stated purpose.
Anthropic accused Alibaba’s Qwen lab of running a large-scale distillation attack on Claude: 25,000 fraudulent accounts generating 28.8 million exchanges between April and June. Alibaba denied it. YourNewsClub maps those disputes as the correct frame for the ban – one episode in a bilateral technology relationship that has fully broken down.
The Alibaba ban does not resolve the underlying disagreement about distillation – Anthropic and other frontier labs prohibit it; Chinese AI labs view the restriction as a US attempt to prevent technology diffusion – but it makes that disagreement more visible.
Owen Radner, who models digital infrastructure as energy-information transport systems, draws the supply chain implication: “Alibaba’s ban formalises a decoupling already underway at the technical level. Every time a Chinese technology company replaces a US AI tool with a domestic alternative, it accelerates the divergence between the US and Chinese AI software ecosystems. The decision to ban rather than simply deprioritise signals that Alibaba is treating this as a security boundary rather than a preference.” Alex Reinhardt, who tracks financial systems and settlement infrastructure through digital protocols, places the commercial context: “Anthropic was reimbursing employees for Claude usage, making it an officially sanctioned vendor. The transition from sanctioned vendor to banned software in a single notice, triggered by a security disclosure, is the kind of event that enterprise sales teams cite when they ask why AI companies need to maintain compliance documentation that covers covert detection mechanisms as well as stated data handling policies.”
Your News Club notes the July 10 effective date as the first concrete benchmark against which Alibaba’s actual Qoder transition can be measured, since internal notices and employee behaviour are not always the same thing in large organisations.
Qoder’s actual capability relative to Claude Code will determine whether this ban sticks operationally. Claude Code’s agentic features – operating autonomously across a codebase, running tests, fixing its own errors – have been cited by developers as qualitatively different from earlier coding assistants. Whether Qoder has closed that gap is a question Alibaba’s own engineers will answer over the next month.
The broader trajectory is the more significant story. China’s leading technology companies view US frontier AI tools as carrying legal, security, and operational risks that domestic alternatives increasingly do not. YourNewsClub calls the accumulation of those individual decisions – Alibaba today, other major Chinese firms tomorrow – as the most commercially concrete measure of US-China AI decoupling at the developer tool level, which is several layers of abstraction below the semiconductor restrictions that dominate public discussion of the same phenomenon.