Twenty organisations including Huawei, Alibaba, Tencent, Xiaomi and Baidu draft an AI agent self-discipline convention
The convention calls for stronger agent governance, compliance-by-design technical measures, and joint resistance to algorithmic exploitation and grey-market abuse through agents.
On November 4 the relevant forum at the 2026 World Internet Conference Wuzhen Summit formally released the AI Agent Development Self-Discipline Convention. The convention was jointly drafted by 20 organisations including Huawei, Alibaba, Tencent, Xiaomi, Baidu and Volcano Engine, and states that agent governance mechanisms should be strengthened, that compliant technical measures should be used to ensure agent behaviour is lawful and legitimate, and that the industry should jointly resist algorithmic exploitation and grey-market activity carried out through agents, while promoting inclusive sharing and open development of agent technology to help build a healthy and constructive development order.
The convention was released against a backdrop of scaling challenges as agents move quickly from conversation to task execution. The industry broadly holds that the internet is shifting from information interconnection and service interconnection toward agent interconnection, and that as governance objects shift from content to behaviour, a unified foundation has become the shared key to faster cross-domain agent collaboration. Three obstacles remain: whether users can know about and promptly revoke agent permissions, whether service providers can confirm which agent was invoked and on whose behalf it acted, and whether agents on different platforms can find and call each other.
Ant Group vice president Chen Liang said at the forum that the agent era requires trusted infrastructure so that identity, authorisation and accountability enter every connection, noting this is both a matter of enterprise product competitiveness and a public infrastructure question for the intelligent internet. At the national level, parties are jointly building an agent identity registration and resolution platform addressing identity encoding, cross-domain resolution and trusted interoperability, with operating platforms on top to carry agent access and capability publishing, and a pilot verification programme advancing steadily.