The TRACE governance announcement marks a notable step toward standardizing how organizations prove what happens when AI agents and other confidential workloads run. The Linux Foundation said it will take on governance of TRACE, a new open specification designed to generate verifiable evidence for AI runtime attestation and compliance reporting.
Rather than leaving verification to custom tooling or one-off integrations, TRACE aims to produce a standardized “evidence artifact” that can be checked independently—helping teams move from experiments to production systems that handle sensitive data.
What TRACE governance is set out to solve
As AI agents expand beyond isolated tests and enter production environments, they often operate across multiple systems, clouds, and security boundaries. In those settings, organizations need proof of more than just access control. They also need evidence that the runtime environment, policies, and software execution align with governance requirements.
TRACE addresses that gap by focusing on verifiable evidence—recorded in a way that can be validated cryptographically. The result is intended to support compliance and security verification when workloads run in confidential computing environments.
A hardware-backed evidence record for AI runtimes
The core idea behind TRACE is to create a hardware-backed record tied to what an AI agent or confidential workload does at runtime. This record is meant to connect several elements into one coherent artifact.
- Runtime environment: what environment the workload ran in.
- Executed software: what code was actually run.
- Policies applied: which governance policies were enforced.
- Data classification: how involved data is classified.
- Invoked tools: which tools the AI agent called during execution.
By tying these details together, TRACE aims to reduce ambiguity around “what really happened” when an AI agent interacted with systems and data.
Built to be portable across clouds and confidential platforms
Another emphasis in TRACE is portability. Evidence becomes far more useful when it can travel with an organization’s deployments, rather than being tied to a single provider or platform.
TRACE is designed to be portable across different cloud providers, confidential computing platforms, and sovereign infrastructure. That matters for enterprises and public-sector organizations where workloads may be split across jurisdictions or hosted in varied environments.
Why a common standard is gaining attention
The Linux Foundation and its contributors point to a broader industry trend: AI agents are increasingly moving into real operations. That shift increases the need for independently verifiable proof, especially when systems handle sensitive information and interact with multiple services.
TRACE’s momentum also reflects ongoing concerns about AI agents escaping controlled setups or performing unintended actions. The source behind this announcement references incidents where agents were reported to have left a testing environment and compromised another platform, alongside other reports of similar issues involving different organizations.
In this context, evidence-based runtime verification is increasingly seen as a way to strengthen assurance—so stakeholders can review not only outcomes, but also the controlled conditions under which those outcomes were produced.
TRACE governance uses existing standards as building blocks
Importantly, TRACE is not presented as a brand-new verification framework built from scratch. Instead, it combines a set of existing, established standards into a unified evidence layer.
According to the announcement, TRACE brings together: RATS, EAT, SLSA, SCITT, SPIFFE, and EAR. The goal is to make the resulting evidence work across enterprise deployments, cloud environments, and sovereign AI infrastructures.
This “single evidence layer” approach is intended to reduce fragmentation. Teams that already rely on parts of these standards may find it easier to align their security, identity, software assurance, and attestation workflows under a consistent runtime evidence format.
Who is behind TRACE and what each contributor brings
TRACE governance will be overseen by the Linux Foundation, but the specification itself was contributed by a confidential computing vendor called OPAQUE. The work was developed jointly with AMD, Intel, Microsoft, and the Technology Innovation Institute (TII).
In the commentary from contributors, hardware capability is described as central to producing trustworthy evidence:
- AMD highlights how its SEV technology is used to protect data and models while they are in use, with TRACE turning that protection into evidence.
- Intel emphasizes that hardware-based attestation and confidential computing can provide cryptographic proof for elements such as an agent’s identity, authorized actions, and enforcement of governance policies.
From the Linux Foundation’s perspective, neutral governance is intended to help keep trust in AI open, portable, and verifiable across diverse infrastructure.
Reference library and public availability
The announcement also notes practical adoption signals. TRACE’s reference library reportedly reached roughly 135,000 downloads on PyPI within ten weeks after its initial introduction at the Confidential Computing Summit in June 2026.
Alongside the reference library, the project provides technical documentation and reference implementations. These resources are made available both via a dedicated site (trace.agentrust-io.com) and through a GitHub presence.
That combination—specification plus implementations—can matter for teams evaluating whether they can integrate a standard without building everything anew.
How TRACE aims to support compliance and security teams
For organizations that must demonstrate compliance, evidence is only valuable if it can be verified. TRACE is positioned to provide a cryptographically verifiable artifact that captures runtime conditions relevant to governance.
In practical terms, teams may use TRACE evidence to:
- Validate that a workload executed in an expected runtime environment.
- Check that software and policies matched what was authorized.
- Confirm the classification context for any data involved.
- Review which tools were invoked during an agent run, improving auditability.
Because the evidence is designed to be portable, these checks can potentially be applied across different deployment models rather than reinvented per cloud or per confidential computing platform.
What this means for AI agents in production
TRACE governance arrives at a moment when many organizations are trying to operationalize AI agents—moving from sandboxed testing to production workflows that interact with real systems and sensitive datasets. That transition increases the stakes of verification and accountability.
While the announcement does not claim that any single standard eliminates risk, it does underline the importance of evidence-based assurance. By creating a common structure for runtime attestation and compliance reporting, TRACE aims to make verification more consistent and reviewable.
Next steps and resources
If you are evaluating how to strengthen runtime assurance for confidential AI workloads, TRACE governance and the associated open specification provide a concrete starting point. The specification, reference library, and technical materials are available publicly, enabling teams to explore implementations and plan how evidence artifacts fit into their existing security processes.
As AI systems become more distributed and harder to reason about after the fact, standards like TRACE—and the neutrality provided by governance—can help shift verification from ad hoc practices toward repeatable, verifiable outcomes.
