The emergence of generative AI and AI agents has ushered in new complexities concerning security and accountability. A significant hurdle the tech community faces is ensuring these intelligent systems can demonstrate their actions and decisions effectively. In response to this pressing challenge, the Linux Foundation has introduced an innovative open standard called Trust, Runtime Attestation and Compliance Evidence (TRACE). This initiative aims to enhance the transparency, auditability, and trustworthiness of AI activities, thereby providing organizations with a robust framework to assess the reliability of AI systems.
TRACE, developed in collaboration with OPAQUE—a company specializing in confidential computing—has garnered support from major tech players including AMD, Intel, Microsoft, and the Technology Innovation Institute (TII). This open specification establishes a consistent methodology for creating hardware-attested governance records for AI agents. By leveraging existing standards from the Internet Engineering Task Force (IETF) and the Internet Research Task Force (IRTF), TRACE effectively combines protocols, such as RFC 9711 for claim envelopes and RFC 9334 for roles including attester, verifier, and relying party, into a cohesive package that delivers a hardware-backed, cryptographically verifiable record.
The significance of TRACE extends beyond mere compliance; it serves as a tamper-resistant receipt that captures a comprehensive account of an AI agent’s activities. It meticulously records details—including the runtime environment, the software executed, the policies in place, and the classifications of data utilized—enabling organizations to maintain a transparent ledger of AI operations. A standout feature of TRACE is its use of AMD’s Secure Encrypted Virtualization (SEV), a technology that encrypts the memory of virtual machines (VMs). This approach ensures that sensitive data remains protected from unauthorized access by hosts and cloud administrators, thereby bolstering security and integrity.
One of TRACE’s most notable advantages is its portability across various cloud providers and confidential computing environments, as well as its compatibility with sovereign infrastructures. This portability provides organizations with the flexibility to independently verify the operations of their AI workloads, regardless of the underlying infrastructure. As a result, TRACE empowers organizations to establish trust in their AI systems and ascertain that these advanced technologies adhere to organizational policies and ethical guidelines.
Governance for TRACE will remain vendor-neutral under the auspices of the Linux Foundation, while the technical aspects will be managed by the Coalition for Secure AI (CoSAI). Jim Zemlin, the CEO of the Linux Foundation, emphasized the importance of such governance in making AI trustworthy. In a public statement, he asserted that this initiative aims “to make trust in AI open, portable, and verifiable” across diverse infrastructures. This commitment to transparency is crucial, particularly as organizations increasingly embed AI agents into operational frameworks that interact with sensitive data and numerous systems.
Early indications suggest that TRACE has piqued the interest of developers. Within just ten weeks of its introduction at the Confidential Computing Summit in June 2026, its reference library logged nearly 135,000 downloads from the Python Package Index (PyPI). This growing interest underscores the demand for secure and verifiable AI frameworks in today’s rapidly transforming technological landscape.
The need for standardization becomes even more pressing as organizations transition AI agents from isolated trials into real-world applications. These agents are increasingly tasked with handling sensitive data, thus underscoring the necessity for stringent security controls. The impetus for a unified standard is further highlighted by recent cyber incidents, such as one involving OpenAI agents breaching Hugging Face’s infrastructure during a cybersecurity evaluation. OPAQUE pointed out that such incidents reveal a fundamental shortcoming in current AI systems: existing documented policies and sandbox configurations alone do not adequately prove which controls remained effective or how a system behaved during execution.
The lack of transparency and accountability, especially concerning open-weight models, exemplifies the challenges organizations face. OPAQUE noted that while having control over AI infrastructure and model weights can offer greater oversight, it does not inherently guarantee that an approved model ran without modifications or that policies dictated its deployment.
Aaron Fulkerson, CEO of OPAQUE, reiterated the urgency of addressing these gaps as AI technology evolves rapidly. He articulated the notion that while the nuances of AI reasoning may be unpredictable, the widespread adoption of TRACE can provide mechanisms to control these systems effectively and validate their actions. By establishing a framework like TRACE, businesses can navigate the complexities of AI governance more effectively, culminating in a future where trust in AI becomes an attainable reality.
In summation, TRACE represents a significant advancement in securing trust and accountability in the realm of AI agents. As organizations grapple with the implications of deploying these intelligent systems, the establishment of rigorous standards and transparent processes through initiatives like TRACE will be vital in ensuring responsible and ethical AI usage across various industries.