In this collaboration, both parties will jointly explore the construction of an AI Agent security certification system, gradually refining corresponding evaluation standards and certification mechanisms based on the application scenarios, security capabilities, and risk characteristics of different types of Agents.

Article author and source: ME News



Context

As AI agents evolve from demos to real-world applications, security concerns have become the industry's top priority. Unlike traditional Web3 products, AI agents can invoke tools, access data, manage wallets, and execute transactions—making the potential impact of any security breach significantly greater.

SlowMist, as a globally leading blockchain security company, has long been dedicated to areas such as smart contract security, wallet security, and threat intelligence, while continuously monitoring new security challenges in the AI Agent era.

AgentOn is dedicated to building an open AI Agent marketplace that enables the creation, discovery, and application of AI Agents, providing ecosystem services for developers and users.

This strategic partnership brings together Moushu’s practical experience in security defense, risk detection, and threat research, with AgentOn’s capabilities in building the AI Agent ecosystem, to jointly explore security infrastructure tailored for the development of AI Agents.

Collaboration Details

As the application scenarios of AI agents continue to expand, their attack surface has gradually evolved from isolated model risks to encompass the infrastructure layer, protocol/tool layer, behavioral layer, and model layer. Due to interactions between these layers, threats propagate and amplify across them.

Therefore, AI Agent security detection must cover the entire lifecycle—from development and deployment to operation—rather than being limited to a single stage.

In this collaboration, both parties will jointly explore the development of security capabilities throughout the AI Agent lifecycle, covering key stages such as pre-deployment detection, runtime risk governance, and continuous security monitoring.

Manswu will leverage its expertise in Web3 security, threat intelligence, and risk detection to conduct an in-depth analysis of the core attack surfaces of AI Agents. Building upon the OWASP Top 10 for Agentic Applications, it will develop risk identification capabilities tailored to AI Agent scenarios, with a focus on critical security risks including agent goal hijacking, tool misuse and exploitation, identity and privilege abuse, AI Agent supply chain vulnerabilities, unintended code execution, memory and context poisoning, insecure inter-agent communication, cascading failures, exploitation of human-agent trust, and malicious agents.

Meanwhile, SlowMist will provide AI Agent security detection capabilities across six dimensions—credential and key security, Prompt Injection protection, dangerous command and code execution prevention, data leakage protection, permission and supply chain security, and auditing and observability—leveraging real-world attack scenarios and security research experience. This will help developers in the AgentOn ecosystem identify potential security risks and integrate security measures throughout the entire lifecycle of AI Agent creation, deployment, and operation.

As the AI Agent ecosystem continues to evolve, establishing transparent and reliable security evaluation and certification mechanisms will become a vital foundation for promoting the healthy development of the industry.

In this collaboration, both parties will jointly explore the construction of an AI Agent security certification system, gradually refining corresponding evaluation standards and certification mechanisms based on the application scenarios, security capabilities, and risk characteristics of different types of Agents.

Through the exploration of a security certification system, we help developers demonstrate the security capabilities of their agents, while enabling users to more easily identify trustworthy agents with robust security features. Certification standards will be continuously refined based on different agent types and real-world application needs, rather than established as a one-time uniform set of guidelines.

In addition to building technical capabilities, both parties will collaborate on promoting the AI Agent security ecosystem, industry engagement, and developer community development, including jointly releasing AI Agent Security Best Practices, hosting technical sharing and exchange events, and participating in industry conferences and developer community activities.

Through sharing security practices, industry collaboration, and ecosystem partnerships, both parties will jointly advance the development of AI Agent security concepts and promote the further implementation of security capabilities within the AI Agent ecosystem.

About AgentOn

AgentOn is an AI Agent-powered earning platform incubated by ME Group, dedicated to building a task collaboration network where AI Agents and humans work together, enabling Agents to operate autonomously, complete tasks, and earn real on-chain rewards.

About SlowMist

Mantou Tech is a threat intelligence company focused on blockchain ecosystem security, founded in January 2018 by a team with over a decade of hands-on experience in cybersecurity defense and offense. The team previously developed security solutions with global influence. Mantou Tech is now a leading international blockchain security company, offering AI-driven, end-to-end security solutions—from threat detection to defense—tailored to specific environments. It serves hundreds of top-tier and well-known projects worldwide, with commercial clients spanning more than a dozen major countries and regions.