China's MIIT moves to draft AI agent interface and MCP security standards
China's Ministry of Industry and Information Technology has been soliciting public comment on its seventh batch of 2026 industry-standard revision plans, which include projects on AI safety governance, security requirements for Model Context Protocol (MCP) applications, and classification and basic requirements for agent interfaces. The report argues that as agents plug into enterprise workflows, external tools and business data, AI safety governance is shifting from the model itself to interfaces, permissions and call chains, where missing boundaries amplify risks of data leakage, unauthorized calls and misoperation. It cites Tianyancha data saying China has more than 210,000 AI-security-related companies, with about 31,500 newly registered so far in 2026.
If the standards take effect, model vendors would have to state capability boundaries, app developers would need permission controls and audit trails, and enterprise customers would have to rework data access rules, making security a design prerequisite rather than a post-launch patch.
The source text is unavailable here. Read the original for the full story.
#Ministry of Industry and Information Technology (MIIT)#Model Context Protocol#Tianyancha