Why did Alibaba completely ban Claude Code?

Amid the wave of accelerated evolution in AI programming tools, an internal adjustment centered on overseas tool security and the pursuit of independent substitution is emerging within major domestic internet companies.

On July 3, a reporter learned from an Alibaba insider that, due to the recent disclosure of a security risk involving an implanted backdoor in Claude Code, Alibaba—after comprehensive evaluation—has included it in its high-risk software list. Starting July 10, Alibaba will completely ban internal employees from using Claude Code in workplace environments and recommends using Qoder as an alternative.

Angel investor and senior AI expert Guo Tao analyzed for a Yicai reporter that Alibaba’s company-wide ban on the Claude series products is a strategic decision that balances data security, compliance risk control, and autonomous transformation, rather than a simple choice of technical tools.

Guo Tao further said that previously, Alibaba internally encouraged employees to use AI tools both at home and abroad, reimbursing in large amounts the costs of external models. Claude Code, in particular, had been a high-frequency coding tool for programmers. But Anthropic’s unilateral accusation that Alibaba carried out model distillation attacks, together with its targeted labeling and banning of Chinese users, refusal to refund, and blocked appeal channels, has exposed Alibaba’s R&D data to risks of leakage and lack of traceability. Alibaba’s internal ban on Claude Code is based on multiple real-world considerations.

“As regulatory factors continue to affect the AI industry, more vendors and enterprises will reassess their reliance on overseas models and tools.” Forrester Vice President and Chief Analyst Dai Kun also told reporters that this will accelerate the penetration of homegrown AI coding tools into large enterprises and continuously drive the formation of a more independent model and development tool ecosystem across China and the broader Asia-Pacific market.

One trend is that industry competition and external pressures among global AI leaders are continuously intensifying, affecting domestic vendors to actively or passively reduce their use of overseas AI tools. On the other hand, in fields such as AI programming, the evolution of domestic AI tools over the past year has also been accelerating. Although there may still be gaps compared with overseas top-tier tools, from the perspectives of both data security and commercialization, the opportunity for Chinese vendors to seize this market is also arriving.

Beyond Alibaba, more companies—including ByteDance and Tencent—also have internal adjustments regarding the use of Claude Code and are waiting to see how things will play out.

“This may be the beginning of a wave of overseas AI tool risk-control rectifications initiated by domestic enterprises. Internet and technology companies will prioritize building an internal, controllable domestic AI R&D system. On the industry side, it will speed up the process of domestic substitution for AI coding tools, and self-developed tools such as Alibaba Tongyi Qoder will gain large-scale deployment scenarios.” Guo Tao said. As the Sino-US AI technology contest sinks down into the domain of underlying development tools, it may end the era in which domestic leading companies used overseas AI coding products without constraints, and shift the AI R&D track from competing on traffic-application use cases to competing with hard-core capabilities centered on security, autonomy, and ecological controllability.

One detail is that while disabling Claude Code, Alibaba also recommended Qoder to its internal employees as an alternative. This is Alibaba’s own agent programming platform. Qoder went live in August 2025. As of May 2026, Alibaba has announced that Qoder has accumulated over 5 million users globally and has completed the end-to-end autonomous development process from requirement analysis to code deployment. Over the past year, AI programming has also been an important market for Alibaba. In May this year, Alibaba’s Qwen3.7-Max model also emphasized breakthroughs in capabilities such as programming and inference.

Another hidden thread behind Alibaba’s ban on Claude Code is the significant impact of AI Coding on cloud vendors’ token revenue. Liu Weiguang, Senior Vice President of Alibaba Cloud Intelligence Group and President of the Public Cloud Business Division, revealed that in his view, within token services, from the perspective of future potential, there is also very large room for markets such as advertising, media, film and television, and short videos. However, compared with large language models in the Coding and Agent direction, they are simply not in the same league.

“Coding is our most important direction. It’s almost for everything.” Liu Weiguang said bluntly. In the previous era of cloud computing, one common difficulty in work was tallying clients’ IT budgets and finding that some of the budget could not be accessed—specifically, software development within client companies and human outsourcing. “Now it’s exactly the opposite. These parts are precisely the 100% hit points of AI coding.”

This means that amid this dispute of mutual bans, there may also be ambitions for domestic tools to seek a “rising” window period.

Guo Tao believes that domestic AI coding tools have already completed localization of basic capabilities and are in a crucial middle stage characterized by benchmarking basic capabilities, tackling high-end engineering challenges, and continuously filling in ecological shortcomings. Basic code completion, bug fixing, and routine project development capabilities have aligned with international levels, but gaps may still exist in some tasks. In addition, supporting ecosystems such as tool adaptation and private deployment still need to be improved.

Alibaba’s Qoder, ByteDance’s Trae, and others are further stepping onto the table, but whether they can push the domestic AI programming track to a new turning point remains a long-term challenge.

“Large enterprises have massive amounts of real engineering data, mature cloud services, and R&D pipelines, enabling them to rapidly iterate on self-developed coding tools, seize the government-enterprise and internet markets, and build an integrated cloud-based AI R&D ecosystem.” Guo Tao said. However, the challenge for domestic vendors lies in the need to keep refining high-end complex engineering capabilities over the long term, the high cost of computing resources, and the problem of fragmented industry ecosystems that urgently require unified optimization.

Dai Kun believes that domestic AI Coding tools are moving from single-point development assistance toward the enterprise-level implementation stage. The challenge is that they need to prove that continuously invested Token costs can be converted into measurable efficiency improvements and business ROI. At the same time, the biggest opportunity for domestic AI Coding in the future is not only replacing code generation tools, but also driving the intelligent-agent transformation of enterprise business processes and existing applications, as well as developing a new generation of intelligent agent applications driven by enterprise knowledge.