Summary
- TRACE has moved under Linux Foundation governance after being contributed by confidential-computing company OPAQUE.
- The specification is designed to create portable, cryptographically verifiable evidence from AI and confidential workloads.
- Its backers include AMD, Intel, Microsoft, OPAQUE, and the Technology Innovation Institute.
The Linux Foundation has taken governance of a new open specification intended to produce cryptographically verifiable evidence about what AI workloads did while they were running.
TRACE — Trust, Runtime Attestation and Compliance Evidence — was contributed by confidential-computing company OPAQUE and has been developed with support from AMD, Intel, Microsoft, OPAQUE, and the Technology Innovation Institute.
The specification is designed around hardware-attested runtime and compliance evidence for AI agents and confidential workloads. The Linux Foundation will provide vendor-neutral governance as the project develops.
The technical problem TRACE is trying to address sits between traditional logging and hardware trust. Application logs can record what software says happened, while hardware-backed attestation can establish properties about the environment in which code ran. Combining those concepts is intended to create evidence that can be verified independently rather than accepted purely because a workload or platform produced it.
That distinction becomes more relevant as AI systems are given authority to call tools, retrieve data, invoke external services, or take actions without every step being manually approved. Governance increasingly depends not only on defining what an AI agent is permitted to do, but on establishing afterwards whether the runtime environment and the resulting actions matched those controls.
Conventional audit trails are useful but can become difficult to rely on when evidence is fragmented across model providers, orchestration layers, cloud infrastructure, plugins, and enterprise applications. A portable attestation model could give different parts of that stack a common way to present evidence about execution and policy enforcement.
TRACE is not itself proof that AI agents can be made safe or compliant. Attestation can establish particular facts about software, hardware, or execution conditions, but it does not determine whether the underlying business rule was sensible, whether an authorised action produced an acceptable outcome, or whether a model’s decision was correct.
Its value will therefore depend on what evidence implementations actually generate, which components can be trusted to produce it, and whether independent systems can validate that evidence across heterogeneous infrastructure. Open governance may help reduce the risk that runtime assurance becomes tied to a single cloud, processor, or AI platform, but interoperability has to be demonstrated in production rather than assumed from a specification.
The involvement of AMD and Intel gives the project a direct connection to hardware trust, while Microsoft adds a major cloud and enterprise software supplier to the initial group. OPAQUE’s work centres on confidential AI and data environments, where proving properties of execution without exposing the underlying workload is already a core architectural problem.
The project arrives as AI governance moves increasingly from policy documents into technical controls. Organisations deploying agents are beginning to confront questions about delegated authority, tool access, evidence retention, and how to reconstruct automated actions after a failure or disputed decision.
TRACE attempts to address one part of that problem: creating evidence that can travel with the workload and be verified rather than relying solely on the word of the system that generated it. Whether it becomes a meaningful assurance layer will depend on adoption, implementation quality, and the extent to which different AI and infrastructure vendors agree on what must be attested.
