At the 2026 World Artificial Intelligence Conference (WAIC) and the High-level Meeting on Global AI Governance held in Shanghai from July 17 to 20, Chen Liang, Vice President of Ant Group and Chief Technology Officer of Ant's Security Division, presented his views.
Chen Liang argued that AI agents are evolving into a new type of operating system for the AI era, necessitating a corresponding foundational framework of trust.
To address this, Ant Group has proposed building the trust foundation for AI agents from two key directions: "native security" and "trusted interconnectivity."
Regarding native security, Ant has developed agent operation platforms adapted to various terminal forms. This includes the LingDevice OS, a wearable device agent operating system designed for lightweight devices like smart glasses and headphones, as well as an enterprise-grade agent operating system for complex business scenarios involving multiple users and tasks.
The wearable device operating system emphasizes device-cloud collaboration, low power consumption, and cross-device connectivity, supporting the operation and distribution of agent services on devices like glasses and headphones. The enterprise-grade system is tailored for complex business environments, providing high-concurrency, distributed scheduling capabilities for multi-agent tasks.
To manage the inherent uncertainties in agent tasks, Ant has integrated native security and a "gear-shifting" orchestration and scheduling mechanism into its agent engine. This approach, supported by a distributed runtime platform, balances execution efficiency, reasoning capability, and operational cost while embedding security functions deeper into the operating system layer.
Chen Liang believes that the traditional "build first, secure later" model is ill-suited to the operational characteristics of AI agents. Security capabilities must be embedded at the operating system level, shifting from patchwork protection to systemic security.
Building on this runtime platform, Ant has also established an agent "immune system" covering evaluation, identity and permissions, runtime protection, and supply chain security.
Ant has proposed the Agent Security Link (ASL) protocol for trusted interconnectivity. ASL can be layered on top of existing agent interoperability protocols. It establishes a verifiable, transferable, enforceable, and auditable trust chain for cross-agent collaboration through four core capabilities: trusted identity, trusted connection, trusted intent, and trusted authorization.
Trusted identity verifies the identity of devices and agents, supporting trusted transfer of identity within the collaboration chain. Trusted connection establishes a secure communication link via end-to-end encryption. Trusted intent validates the source of instructions to reduce the risk of tampering. Trusted authorization adheres to the principle of least privilege, enforcing authorization boundaries during multi-level delegation.
ASL leverages underlying security technologies such as Trusted Execution Environment (TEE), Decentralized Identifiers (DID), and Public Key Infrastructure (PKI). These capabilities are currently applied in scenarios including smartphone assistants, AI glasses, smart vehicle systems, and embodied intelligence.
Chen Liang clarified that ASL is not intended to replace existing agent connection protocols. Instead, it aims to add a layer of trusted interconnectivity, enabling agents from different entities to achieve verifiable identity, tamper-resistant intent, bounded authorization, and auditable processes beyond mere connection and invocation.
He emphasized that for AI agents to achieve large-scale application, it is essential not only to continuously improve model capabilities but also to complete the infrastructure, including operating systems, collaboration networks, and trust protocols.
Ant Group is collaborating with over 20 partners, including Qwen, Xiaomi, Zhipu, BYD, and SenseTime, to advance the refinement and practical application of the Agent Security Link (ASL) protocol.