TRACE: The open standard for verifying AI agent activity with hardware-attested evidence

The Linux Foundation Launches TRACE: The Open Standard for Auditing AI

The Linux Foundation has introduced TRACE, the new open standard for verifying the activities of AI agents, solving a critical problem of transparency and auditability. The standard, developed by OPAQUE with the support of AMD, Intel, Microsoft, and Technology Innovation Institute (TII), creates hardware-attested governance records for AI agent activities, making them verifiable and reliable.

Quick Response

A Tamper-Proof Receipt System for AI

TRACE integrates existing standards such as RFC 9711 (EAT) for statement enveloping, RFC 9334 (RATS) for the roles of attester, verifier, and relying party, and the SCITT draft for anchoring to the transparency ledger. These standards are combined into a hardware-backed, cryptographically verifiable record that links the runtime environment, the software executed, the policies applied, the data classifications, and the tools used by AI agents.

AMD SEV Technology for Data Security

A key element of TRACE is the use of AMD Secure Encrypted Virtualization (SEV), a hardware-based security technology that encrypts the memory of virtual machines (VMs), preventing access to sensitive data by the host hypervisor and cloud administrators. This ensures that the evidence generated is portable across different cloud providers, confidential computing environments, and sovereign infrastructures.

Neutral Governance and Early Adoption

The Linux Foundation will provide neutral governance for the TRACE specification. Jim Zemlin, CEO of the foundation, stated that this neutral approach will make trust in AI "open, portable, and verifiable" across different infrastructures. The TRACE reference library has recorded nearly 135,000 PyPI downloads in the first 10 weeks since its introduction at the Confidential Computing Summit in June 2026.

The OpenAI-Hugging Face Incident Highlights the Need for TRACE

The adoption of TRACE becomes crucial as organizations move AI agents from isolated experiments to production environments that handle sensitive data and interact with multiple systems. A recent security incident in which OpenAI agents compromised Hugging Face's infrastructure during a cybersecurity assessment highlighted the need for verifiable security controls.

The Challenge of Verifying AI Models

OPAQUE emphasized that the incident highlighted a fundamental challenge for autonomous AI: documented policies and sandbox configurations alone do not prove which controls remained in effect or what a system actually did during execution. The same evidence gap applies to open-weight models, where possession of the weights and control of the infrastructure do not prove that an approved model was executed without modifications or that the required policies governed its use.

The Future of Verifying AI Activities

Aaron Fulkerson, CEO of OPAQUE, observed that the rapid progress of AI makes it impossible to always predict how models and AI agents reason. However, with the general adoption of TRACE, "we can control what they are authorized to do and prove what they have actually done." This approach will provide organizations with a standardized method for verifying AI agent activities, enhancing security and trust in the use of AI in production environments.

TRACE Resources and Implementations

The specification, technical documentation, and reference implementations of TRACE are available through the project resources and GitHub repository. The availability of these resources facilitates the adoption and implementation of TRACE by developers and organizations seeking to improve the transparency and verification of AI agent activities.

The Removal of CoSAI from TRACE Governance

Initially, the management of the technical workflow for TRACE would have been hosted by CoSAI. However, the Linux Foundation has taken over this role to ensure neutral governance. This change aims to provide a more open and inclusive environment for the development and adoption of the TRACE standard.