At the 2026 Yunqi Conference on the afternoon of September 23, AntM密算 Chairman Wei Tao unveiled the company's trusted intelligence strategy and product roadmap for the first time.

According to Wei Tao, AntM密算 is building its trusted intelligence capabilities around a three-tier architecture: a "dual-foundation—Fabric framework—密算 service platform." The bottom layer consists of two pillars: confidential computing power and the HOP (High-Order Program) trusted native intelligent agent foundation. The middle layer utilizes a trusted intelligence Fabric to organize intelligent agents, knowledge, data, and security capabilities into a complete architecture. The top-tier 密算一号 platform delivers data governance, query analysis, analytical modeling, and industry-specific intelligent services to enterprises.

In Wei Tao's view, over the past two years, most enterprises' AI initiatives have remained in a "trial phase"—models perform impressively in demonstrations and pilot projects but consistently fail to take over real business operations. The reason lies in the fact that general-purpose models can grasp public knowledge, yet processing an insurance claim or verifying a transaction still requires specific contracts, up-to-date records, and applicable rules. Completing tasks also demands that human authorization be fully honored, critical steps not be skipped, and execution results be verifiable and traceable.

What enterprises need is not merely a smarter model, but a set of trustworthy production capabilities. Consequently, AntM密算 proposes the concept of "trusted intelligence," which goes beyond preventing data leakage to include trustworthy use of computing power, reliable execution of intelligent agents, controllable and auditable business processes, and continuous accumulation of specialized knowledge.

Regarding confidential computing power, AntM密算 focuses on enabling enterprises to securely utilize cloud-based resources. Confidential computing integrates cryptography, trusted hardware, and system security, allowing enterprises to leverage external computing power, models, and intelligent agent services without relinquishing control over their data. Wei Tao noted that as enterprises feed data into AI systems, the protection scope of confidential computing has expanded to cover prompts, knowledge bases, and the context, memory, and intermediate results generated during intelligent agent operations—ensuring these elements participate in processing and reasoning within protected environments and deliver outputs as agreed, eliminating the risk of platform administrators or model service providers accessing enterprise data and intelligent agent assets without authorization.

In terms of intelligent agent execution, AntM密算 employs HOP to enforce task constraints and execution verification. Wei Tao believes that when intelligent agents can autonomously plan, invoke tools, and operate systems, enterprises face three new types of loss-of-control risks: intent alignment—where human authorization and intentions may not be accurately understood or executed by AI; execution control—where complex tasks may encounter loop execution or deviate from objectives; and asset protection—where data and knowledge may leak during invocation and transmission. HOP, as a native language for intelligent agents, structures task objectives, exploration, inspection, and submission steps into protocols that are easy for both humans and agents to understand and align with, upgrading intelligent agents from merely being able to complete tasks to being able to complete them reliably and trustworthily.