Agentic Trust FrameworkIndependent Implementations Converged on the Same Five Elements.
The strongest evidence for an architecture is other people arriving at it without being told to. Two independent implementations map onto ATF's five elements, one of them from Microsoft.
Microsoft Agent Governance Toolkit
A seven-package open-source toolkit providing runtime security governance for AI agents. Launched 2 April 2026 under the Microsoft open-source organization.
Integrations span LangChain, AutoGen, CrewAI, the OpenAI Agents SDK, Google ADK, and more. The toolkit team has engaged directly on ATF conformance via GitHub.
| Toolkit package | ATF element | What it does |
|---|---|---|
| Agent Mesh | Identity | DID-based agent identity with behavioral trust scoring |
| Agent OS | Behavior | Stateless policy engine with sub-millisecond enforcement |
| Agent Compliance | Data Governance | Automated governance verification and regulatory framework mapping |
| Agent Runtime | Segmentation | Dynamic execution rings with capability sandboxing |
| Agent SRE | Incident Response | SLOs, circuit breakers, kill switches |
Two further packages sit outside the five-element mapping: Agent Marketplace (plugin supply-chain security) and Agent Lightning (reinforcement-learning training governance).
Berlin AI Labs
A 12-service reference implementation covering all five ATF elements, with contract validation testing across each one.
An independent implementation demonstrating that ATF applies cleanly to microservices-based agent architectures, not just monolithic ones.
Nobody coordinated this.
Two teams, working independently, built agent governance and landed on the same five concerns. That is what it looks like when an architecture is describing something real rather than inventing a taxonomy.
Building on ATF?
The specification is open, and implementations are welcome. If you have built against ATF, or want conformance reviewed, open an issue on GitHub or get in touch.