"Sun Wukong" is born! Alibaba establishes the ATH Business Group

On March 17, just one day after Alibaba announced the establishment of the Alibaba Token Hub (ATH) business group, the enterprise-grade AI-native work platform “Wukong,” built by the DingTalk team, officially debuted.

Wukong is a standalone application, aiming to give every team and every company a “fleet” of AI Agents that work 24 hours a day. Starting today, Wukong opens for invite-only testing and will also be directly built into DingTalk for more than 20 million enterprise organizations.

According to information, the ATH business group is directly led by the Group CEO Wu Yongming, and its core mission is to integrate AI business around Token as the centerpiece—building a complete AI productivity ecosystem covering the creation, delivery, and application of Tokens. This organizational structure adjustment also marks a new stage in Alibaba’s artificial intelligence strategy.

The first layer is a dual-layer rules system. Basic security rules are the bottom line for AI behavior. AI will not sacrifice data security and user privacy to complete tasks. On this foundation, enterprises can define custom rules so that the security framework truly matches each company’s real needs.

The second layer is unified identity authentication—every Wukong user in an enterprise must log in through the company account. The system ensures security through a three-tier mechanism: device binding, horizontal permission control, and user confirmation. Wukong’s permissions are the intersection of the user’s permissions and the questioner’s permissions. In terms of data security, when the content generated by AI contains sensitive data, the system automatically replaces it with a de-identification marker, and sensitive information will never reach users.

The third layer is a dedicated sandbox. Wukong uses container-level sandbox isolation and follows the principle of least privilege. Even if a certain Skill has a vulnerability, the attack surface is strictly limited to a single sandbox and will not affect other Agents or enterprise data.

The fourth layer is the Skill ecosystem. All Skills, regardless of source, automatically undergo security scanning before being listed. During runtime, the Policy engine evaluates every step in real time; any operation that does not comply with the policy is immediately blocked.

第五层即专属模型。对于金融、医疗、政务等行业，可以将模型部署在自有环境中，数据不出企业边界，同时享受悟空完整的Agent能力框架。

第六层即网络代理。所有网络请求可追溯可管控。企业不仅知道谁发了请求，还知道从哪台设备、通过哪个进程、在什么环境下发出。网络代理确保所有被放行的流量都带有完整身份信息，实现放行也可追溯。