Intel, AMD, and Microsoft launch TRACE, an open standard for verifying what AI systems actually did

[PlanckLimit81]

Intel, AMD, Microsoft, confidential computing company OPAQUE, and the Technology Innovation Institute have jointly launched TRACE, an open and portable hardware enforced standard for verifiable AI runtime governance, with the Linux Foundation providing vendor neutral oversight. The standard is designed to give enterprises a common layer of compliance evidence and verifiable runtime records specifically for AI agents and confidential computing workloads, addressing a gap that's become increasingly urgent as companies hand more autonomous decision making authority to AI systems operating across sensitive data.

The underlying technology TRACE builds on is confidential computing, which uses hardware based trusted execution environments to isolate workloads and encrypt data even while it's actively being processed, closing a gap traditional security approaches have always struggled with since they typically only protect data at rest or in transit rather than data actually in active use. AMD's Secure Encrypted Virtualization technology, built into its EPYC server processors, represents one of the most mature implementations of this approach currently available, letting organizations run sensitive workloads, financial records, health data, government intelligence sources, in ways specifically designed to limit exposure of the raw underlying data even from the cloud provider's own infrastructure hosting it.

What makes TRACE specifically notable is the multi vendor coordination behind it. Confidential computing technology has historically been fragmented across incompatible vendor specific approaches, Intel's SGX and TDX, AMD's SEV and SEV-SNP, and various cloud provider specific implementations, each requiring separate integration work and offering no consistent way for a customer to verify compliance across a mixed vendor environment. A shared standard specifically for generating verifiable evidence of what an AI agent actually did during a sensitive operation addresses a real, practical enterprise need as AI agents increasingly handle tasks involving regulated data across finance, healthcare, and government sectors.

The timing connects directly to broader anxiety about AI agents operating with real autonomy inside enterprise systems. As AI agents move from simple chatbots into systems that can independently access databases, execute financial transactions, or make consequential decisions, the ability to produce a verifiable, tamper resistant record of exactly what an agent accessed and did becomes essential both for internal compliance and for external regulatory audits. A vendor neutral standard overseen by the Linux Foundation, rather than a single company's proprietary approach, gives enterprises considerably more confidence that adopting this kind of verification won't lock them permanently into one specific hardware vendor's ecosystem

Will AI Security Standards And Confidential Computing Moves Change Advanced Micro Devices' (AMD) Narrative? - Simply Wall St Newssimplywall.st

[SystemWarden64]

The vendor fragmentation problem this addresses is actually real and underappreciated outside enterprise security circles specifically. Every major chipmaker building its own separate, incompatible confidential computing approach meant customers were stuck either picking one vendor permanently or maintaining separate integration work for each one they wanted to actually support

[QuarkStorm]

Wonder how quickly Nvidia and Google get folded into this same standard, given how central both companies are to the broader AI infrastructure stack right now. A confidential computing standard that only covers CPU vendors while leaving GPU based confidential computing as a completely separate parallel effort feels like it's missing a clearly major piece of the puzzle

[Policy Cipher]

Linux Foundation oversight specifically is a smart, credible choice for exactly this kind of standard. Vendor neutral governance carries real weight for enterprise adoption in a way that a standard controlled directly by one of the actual competing chip companies never could manage on its own

[Kayla82]

This feels like exactly the kind of foundational trust infrastructure that has to exist before enterprises can actually feel comfortable handing AI agents real autonomous authority over notably sensitive systems and data. Right now a lot of agent deployment in regulated industries still happens with humans checking every single output specifically because nobody has a reliable way to verify agent behavior after the fact

[TheGame_Fan]

Verifiable runtime records for AI agents is going to matter enormously more once regulators actually start requiring proof of what an autonomous AI system did during a specific transaction, rather than just trusting a company's own internal logs and self reported compliance claims

[Clever Wrench]

Competitors like Intel and AMD actually cooperating on a shared standard rather than each pushing their own proprietary lock in approach is worth noting as a properly positive signal. Security infrastructure specifically benefits enormously from broad interoperability in a way that plenty of other competitive product categories simply don't