Shanghai Issues Digital ID Codes to AI Agents as Firms Redraw Work

At the event, the Shanghai AI Industry Association and the China Electronics Standardization Institute, among others, promoted AIP agent-interconnection applications. Staff described the identity code as a “digital ID” and walked through the process from application and material verification to cross-entity invocation. Under a series of national standards for agent interconnection, one identity code corresponds precisely to one agent. For organizations, the code is comparable to an employee number, but the report said the number is only the starting point.

The more practical question is who assigns work to agents, who accepts their results and who bears responsibility when something goes wrong. As agents move from tools on personal computers to digital co-workers that call across systems and participate in business processes, the report said organizations must redraw their division of labor. The shift is from “one person, one tool” to a situation in which one person may have a team of agents behind them.

Policy guidance has been moving in parallel. On July 17, at the 2026 World Artificial Intelligence Conference in Shanghai, the Cyberspace Administration of China and relevant parties proposed a global cooperation initiative on mutual trust, interconnection and interoperability of agents, calling for more research investment in basic technologies such as agent identity identifiers and collaborative mutual recognition. The International AI Ethics Governance Action Plan, released at the same time, proposed differentiated risk rules around factors including tool-call permissions and autonomous decision-making capacity, and defined responsibility boundaries for developers, deployers and users. The Implementation Opinions on Standardized Application and Innovative Development of Agents, issued by three departments including the CAC, proposed exploring an agent registration platform that provides digital identity management, retrieval and discovery, and capability declaration services. Shanghai’s “Digital Shanghai” 15th Five-Year Plan, released in August, called for building a government-agent cockpit to observe operations and promote efficient collaboration among people, agents and systems.

Company pilots show how the division of work is being tested. SAIC Fanyi Shangxing has put more than 100 digital employees online. After its planning and scheduling assistant produces simulated plans, the plans are handed to planners for review. Chen Ning, head of R&D for the digital employee project, said the essence of a digital employee is a virtual collaborator that can undertake complete job responsibilities and focus on a single position to achieve a full-process closed loop. SenseTime’s Xiaohuanxiong lets business staff complete data association, trend analysis and report generation by entering a natural-language sentence, freeing people from data retrieval to focus on decisions. The report said the key is not the number of assistants but whether each task has a clear owner, including which links go to agents and which must be checked by humans.

Organization forms are changing with the technology. Employees can carry multiple agents, and staff can form task-based teams that dissolve and regroup when work is done. Managers take on the added job of designing human-agent division of labor and backing the results. In a traditional hierarchy, tasks pass down through departments and ranks; matrix organizations allow cross-department work but still rely on relatively fixed reporting lines. The report described a further shift toward tasks and roles, in which an employee can dispatch multiple agents and professionals form temporary project teams around a task. Assigning work should be led by people who understand the business. Guotai Haitong aims to use AI to drive organizational structure change, business-process innovation and business-model upgrading; its AI applications cover nine business areas and 260 sub-scenarios, with more than 1,000 business agents deployed and more than 500 employees involved in building them, including more than 100 frontline business staff. Acceptance needs rules. People’s Daily Online’s No. 001 AI digital employee is assigned as an “Office Digital Collaboration Specialist,” with an assessment system that sets authorization red lines and uses task completion, acceptance pass rate and operational compliance as core indicators. Final decisions remain with people: the implementation opinions clarify the boundaries among decisions reserved for users, authorized decisions and autonomous agent decisions, and ensure users’ right to know and final decision-making power.

Accountability becomes more complex when several agents relay a task. One agent may initiate a task, another may call a system, and a third may execute an operation; if an unauthorized action or error occurs, companies must determine which link to trace. That requires identity authentication and permission controls, as well as task logs, operation records and human approval nodes so that each delegation has a source, each operation has a boundary and each result has a responsible entity.

From issuing identity codes to defining responsibility for digital employees and rebuilding human-agent management mechanisms, each step is creating demand for professional services. Consulting and planning, process reengineering, system integration, operations management and security governance are expected to form a service chain for scaled agent applications, in line with the Ministry of Industry and Information Technology’s special action to cultivate AI application service providers. The 2026 Global Developer Pioneer Conference is scheduled for Oct. 23-25 at Shanghai Zhangjiang Science Hall. Giving agents “employee numbers” is only the first step; the next is to answer an old management question: how people divide work, how tasks are coordinated and how responsibility is enforced.

Editor's Summary

Shanghai has begun issuing digital identity codes to AI agents, giving them traceable identities and raising questions about task assignment, approval and accountability. Policy guidance and corporate pilots are building identity, authorization and human oversight into agent deployment. The next step is to turn agent “employee numbers” into workable management rules.