ValidMind has open-sourced Atryum, a governance layer for AI agents that evaluates autonomous actions rather than just access permissions, aiming to provide a common control framework across agent ecosystems.
ValidMind has launched Atryum, an open source control layer for AI agents designed to provide governance, oversight, and control over autonomous actions. Available now on GitHub, the project aims to address a growing gap in agentic AI as agents increasingly gain the ability to move money, write to production systems, and update records without human intervention.
Unlike conventional security tools that verify credentials and permissions, Atryum evaluates whether an action itself is appropriate for an agent’s assigned role and authority. The platform sits directly in the execution path of AI agents, intercepting tool calls, evaluating actions against policies, routing decisions to humans when necessary, and creating auditable records of every decision.
The framework operates across protocol, harness, and platform layers and is both runtime-agnostic and model-agnostic, enabling organisations to apply governance consistently across different AI models and agent platforms. ValidMind positions Atryum as a shared governance foundation that developers and platform teams can build upon rather than creating separate governance systems for every agent framework.
Alongside the open-source release, the company has opened early access to ValidMind Agent Authority, an enterprise offering built on Atryum. The product will add capabilities such as LLM-as-judge policy evaluation, user- and group-based approval routing, agent-specific policy hierarchies, enterprise IAM integration, and audit analytics.
“Financial institutions are about to inherit a workforce they have never learned to manage,” said Jonas Jacobi, Co-Founder and CEO of ValidMind. “So they hobble it: every decision is routed to a human, or agents are restricted to the point of uselessness. But you don’t capture the value of an agentic workforce by holding it back; you capture it by governing well enough to let agents act.”
