TRACE, short for Trust, Runtime Attestation and Compliance Evidence, creates cryptographically verifiable records showing what software executed, the policies governing it, the class of data it accessed and the tools it called. The initiative is backed by AMD, Intel, Microsoft, confidential-computing company OPAQUE and Abu Dhabi’s Technology Innovation Institute.
OPAQUE contributed the specification to the Linux Foundation, which announced its stewardship on August 25. The move is intended to establish TRACE as an interoperable evidence layer rather than another proprietary security mechanism tied to a particular cloud provider, chip manufacturer or AI platform.
The specification addresses a growing problem created by agentic AI. Conventional audit logs can record actions after they occur, but their reliability ultimately depends on the system producing and storing them. TRACE is designed to create a signed Trust Record inside a trusted execution environment and connect that evidence to hardware attestation, making subsequent alteration or fabrication substantially harder.
A Trust Record can identify the model, code and runtime involved in an execution, the policy applied, relevant data classifications and the tools invoked. The information can then be independently verified by an auditor, customer or another system without relying solely on assurances from the infrastructure operator.
The project has arrived as companies move AI beyond chatbots towards agents capable of executing code, handling email, accessing corporate databases, invoking application programming interfaces and carrying out multi-stage tasks. That transition has raised concerns over credentials, unauthorised tool use, data leakage and the difficulty of reconstructing exactly what an autonomous system did.
Security authorities and industry groups have increasingly focused on these risks during 2026. Work on AI-agent standards has highlighted the need for stronger identity, interoperability and security mechanisms, while experts have warned that conventional cybersecurity controls need adaptation when software can independently make decisions and interact with external systems.
TRACE builds on existing technologies instead of attempting to replace them. Its architecture incorporates established mechanisms covering remote attestation, evidence tokens, software provenance, workload identity and transparency records. This approach is intended to reduce fragmentation and allow evidence generated in one environment to remain usable when workloads move between different hardware platforms, data centres and cloud services.
The underlying security model relies heavily on confidential computing, which protects data while it is being processed rather than only when it is stored or transmitted. Trusted execution environments isolate workloads from other software on the machine and can provide cryptographic measurements demonstrating the state of the hardware and software running inside them.
That capability has become particularly significant for AI as workloads increasingly span CPUs, GPUs and other accelerators. Compromise of one component can potentially expose model weights, prompts, credentials, intermediate data or agent memory unless trust protections extend across the computing chain.
TRACE’s first reference implementation, Confidential MCP, applies runtime attestation and policy enforcement at the Model Context Protocol boundary. The reference work covers technologies including Intel TDX, AMD SEV-SNP and confidential GPU environments associated with Nvidia H100 and Blackwell hardware.
The specification also responds to a compliance challenge confronting organisations deploying AI in regulated sectors. Companies may be required to demonstrate that particular safeguards were operating during a transaction rather than merely showing that appropriate policies existed on paper. Portable, independently verifiable evidence could therefore support audit processes covering financial services, healthcare, government workloads and other environments handling sensitive information.
The project nevertheless remains at an early stage. TRACE version 0.2 is designated a developer preview rather than a production-ready final standard, and its documentation cautions that organisations should examine its limitations before relying on it operationally. Governance provisions and some technical requirements can continue to change before a 1.0 specification is ratified.
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
