Amazon Releases Dogwood: Open-Source Runtime Verification for Safe AI Agents
Key Takeaways
- ▸Dogwood uses temporal logic to enforce policies on sequences of AI agent actions, going beyond point-in-time authorization to regulate workflows and tool-use patterns
- ▸The language is fully compatible with existing Cedar policies, enabling incremental adoption without requiring migration of current deployments
- ▸Released as open source under Apache 2.0 and integrated into Amazon Bedrock AgentCore Policy, Dogwood is immediately available for production use and community contribution
Summary
Amazon has released Dogwood, an open-source governance language designed to address safety risks in AI agents by regulating their tool use through runtime verification. Built on temporal logic and compatible with the existing Cedar language used in Amazon Bedrock's AgentCore Policy, Dogwood adds powerful temporal conditions that evaluate sequences of events rather than just individual requests, enabling teams to enforce complex constraints such as approval prerequisites, rate limits, and action ordering.
The language is released under the Apache 2.0 license and is now integrated into AgentCore Policy, allowing customers to write policies using standard development tools and IDEs. Dogwood solves a critical limitation of point-in-time authorization by looking back over an agent's recent event history, making it possible to express rules about multi-step workflows—for example, requiring approval before acting, staying under running limits, or preventing external communication after accessing confidential data.
This release represents a significant step forward in agentic AI safety, providing teams with mathematical rigor and precise control over agent behavior in production environments.
Editorial Opinion
Amazon's release of Dogwood reflects a mature approach to agentic AI safety—rather than restricting agent capabilities, the company is providing governance tools that let teams define precise, verifiable rules about how agents interact with the world. By grounding the language in temporal logic and making it open source, Amazon is raising the bar for runtime verification across the industry, potentially becoming a standard for organizations deploying high-stakes AI agents.
