Open-source safety initiatives unveiled as AI agents enter everyday business

As artificial intelligence agents gain traction across industries, leading Chinese and overseas institutions have introduced new initiatives to address the risks tied to the large-scale deployment of general-purpose large models.

The Open-Source Safety Initiative for Large Models was launched on Monday at the 2025 World Artificial Intelligence Conference, aiming to strengthen oversight and security in rapidly expanding open-source ecosystems.

The initiative was jointly issued by the China Academy of Information and Communications Technology, in collaboration with the United Nations Industrial Development Organization, the British Standards Institution, Ant Group, and other industry and research partners.

The proposal calls for the global AI community to adopt four key principles: establishing safety standards, securing open-source infrastructure, encouraging responsible AI use, and promoting inclusive international collaboration.

The CAICT also released a new technical report, Terminal AI Agent Safety (2025), the first in China focused on safety protocols for AI agents in end-user and enterprise systems.

The report outlines a multi-level approach to control model behavior, ensure data protection, and define human-AI interaction standards.

Smarter AI agents

It comes as intelligent AI agents are being embedded into everyday business tools – from smartphones to customer service platforms.

One such platform was presented by Beijing-based tech firm Cloopen, which unveiled updates to its Rongxi Agent & Copilot system during the WAIC, which closed on Monday.

It demonstrated how the latest platform helps banks, call centers, and logistics firms automate customer service, analyze sales trends, and respond in real time.

"We've seen companies reduce customer resolution times from days to minutes," said Kong Miao, vice president of Cloopen.

The platform will be used in banking, insurance, and retail sectors across China, and adapted for multilingual support in Southeast Asia to handle large volumes of customer interactions.

Cloopen's product director Tang Xingcai said while performance is important, stability and transparency are equally critical. "An AI agent should provide reliable support, but it must also make its decisions traceable," he added.