If you’ve ever wondered study how to illustrate that an AI agent without a doubt did what it claimed to total, on the hardware it claimed to utilize, with the records policies it claimed to utilize, you’re asking the wonderful quiz. The Linux Foundation acceptable launched this is able to perchance per chance govern an birth regular designed to resolution it.
TRACE, which stands for Trust, Runtime Attestation and Compliance Evidence, is an birth specification that generates hardware-attested, cryptographically verifiable records of AI agent and confidential workload execution. Give it some thought as a tamper-proof receipt for every AI operation, one who a third event can test without needing to rob the operator’s be conscious for the rest.
What TRACE without a doubt does
The specification used to be within the muse developed by OPAQUE, a confidential computing company, in collaboration with AMD, Intel, Microsoft, and the Technology Innovation Institute. Reasonably than constructing a utterly unique framework from scratch, TRACE integrates several established requirements along side RATS (RFC 9334), EAT (RFC 9711), SLSA, SCITT, SPIFFE, and EAR to place definite interoperability across numerous cloud platforms and staunch computing environments.
The core output is what the project calls “Trust Information.” These paperwork capture operational particulars love the runtime atmosphere, the tool that used to be performed, records classifications, and the policies that had been enforced. All of it is secured by silicon attestation mechanisms, that capacity the verification is rooted in hardware reasonably than tool promises.
In intellectual phrases, TRACE leverages hardware attestation applied sciences love AMD SEV and Intel TDX. TRACE in fact offers those capabilities a standardized language to communicate when it involves AI workloads.
The specification used to be first offered on the Confidential Computing Summit on June 23, 2026. In the roughly ten weeks since, its reference implementation has been downloaded approximately 135,000 situations on PyPI, the Python kit repository.
Why neutral governance matters right here
Jim Zemlin, CEO of the Linux Foundation, emphasised the significance of neutral governance for creating verifiable AI have faith records. Ongoing technical pattern will continue below the Coalition for Win AI, identified as CoSAI.
The backing consortium entails AMD and Intel for hardware attestation roots, Microsoft and its Azure confidential computing stack, and TII, primarily based entirely within the UAE.
The peril TRACE is fixing
As organizations deploy increasingly extra independent AI brokers to handle sensitive workloads across numerous infrastructure, a necessary have faith peril emerges. How does a regulated bank test that its AI model ran inside of a staunch enclave on a explicit cloud provider? How does a healthcare company demonstrate to auditors that patient records used to be processed in accordance to coverage? How does any group prove compliance when the workload ran on any person else’s hardware?
TRACE flips the model. Since the attestation originates from the hardware itself, the ensuing Trust Information may per chance perchance even be verified by any third event with entry to the specification. The cryptographic verification can narrate how one thing ran without revealing what it processed.
Disclosure: This article used to be edited by Editorial Team. For added records on how we construct and review bid, stare our Editorial Coverage.
