SlowMist has formed a strategic partnership with AgentOn to develop security capabilities for autonomous AI agents operating across blockchain environments. The collaboration will focus on lifecycle assessments, risk controls and an evolving certification framework, combining SlowMist’s security research with AgentOn’s platform for creating, discovering and using AI agents. The companies have not announced a launch schedule for the resulting tools.
The initiative targets a problem that becomes more consequential as agents gain authority to invoke external tools, access sensitive information, control wallets and execute on-chain actions. A compromised agent can expose more than a single application, because malicious instructions, stolen credentials or unsafe integrations may move through several connected components before affecting funds or smart contracts.
Security Reviews Expand Beyond the Underlying Model
SlowMist and AgentOn plan to examine risks across infrastructure, protocols and tools, agent behavior and the models themselves. Their proposed approach covers the full agent lifecycle, including checks before deployment, controls during operation and continuous monitoring after an agent begins interacting with users and blockchain systems.
The security work will draw on the OWASP Top 10 for Agentic Applications and address threats such as goal manipulation, tool abuse, privilege escalation, supply-chain weaknesses, unintended code execution and poisoned memory or context. SlowMist also identified six practical assessment areas, covering credentials and keys, prompt-injection defenses, dangerous commands, data exfiltration, permissions and supply chains, and auditability.
Certification remains an area for exploration rather than a finished product. The partners said standards would need to reflect the use case, capabilities and risk profile of different agents instead of applying a single static test to every system. Any certification mechanism is therefore expected to evolve, and neither company has specified when assessments, labels or developer-facing security services will become available.
Marketplace Growth Raises the Stakes for Agent Security
AgentOn describes itself as an AI-agent earning platform and marketplace where autonomous systems and people can collaborate on tasks and receive on-chain rewards. For developers in that ecosystem, the partnership is intended to integrate security earlier in agent creation and deployment, rather than treating reviews as a final check after an application is already active.
The agreement also fits a wider push to formalize safeguards around agentic software. SlowMist recently joined Cyberport’s Web4.0 & Agentic AI Security Alliance as a founding member, while Cyberport confirmed the establishment of the group as part of its broader technology ecosystem. Those initiatives suggest security standards are becoming an infrastructure priority, not merely a feature added by individual applications.
Blockchain networks are simultaneously making it easier for agents to reach on-chain services. Hedera, for example, has introduced a hosted Model Context Protocol server and an open Agent Skills marketplace, while keeping transaction signing outside the hosted server. That design illustrates the central security tradeoff, giving agents useful network access without automatically handing them custody or unrestricted transaction authority.
The immediate commitment is to research, assessment design and ecosystem collaboration, including best-practice publications and technical knowledge sharing. The partnership’s impact will depend on whether those plans become usable controls for developers and transparent signals for users, particularly when agents interact with wallets, smart contracts and liquidity venues.
