Written by Silvia Pavelli
The confidential computing industry has spent the last decade solving a hard problem: protecting data while it is actively being used. With technologies like Intel Trust Domain Extensions (TDX) and hardware-enforced execution environments, the battle to secure data in use is largely won. However, the harder and ultimately more important question follows: how does the party relying on a confidential workload verify independently that the workload is actually running where it is supposed to, on hardware in the state it is supposed to be in?
That is the question Intel Trust Authority answers. As of May 18, 2026, it is a built-in capability of every SecretVM, integrating independent hardware attestation directly into the confidential AI inference pipeline.
The Trust Tax in Enterprise AI
Modern enterprise AI adoption is hampered by a significant trust tax. Organizations deploying proprietary large language models or processing sensitive data—such as financial records or healthcare diagnostics—must rely on infrastructure providers to protect their underlying clusters. Even when using Trusted Execution Environments (TEEs) like Intel TDX, the user historically had to trust the operator’s attestation mechanisms.
The integration of Intel Trust Authority into SecretVM changes this dynamic by decoupling roles. Intel acts as the independent appraiser, the host remains strictly the host, and the relying party receives an attestation token signed by an independent third party with no operational stake in the workload. This architectural shift ensures that verification does not require trusting the operator.
What Is Actually Shipping
SCRT Labs has integrated Intel Trust Authority directly into the SecretVM platform with three critical properties that simplify and secure confidential AI inference:
- Attestation is on by default. Every SecretVM now exposes an `/ita_jwt` endpoint that returns an Intel-signed attestation token on demand. There is no separate sign-up flow, no API key to provision, and no SDK to integrate. SCRT Labs ships its own ITA credentials and a Secret-defined appraisal policy bundled into the platform.
- Custom policies are a one-field addition. Operators who want their VMs attested against their own ITA accounts and policies can add API keys directly in the SecretVM configuration. The endpoint then issues tokens against each configured policy, returned as an array.
- Tokens are bound to the TLS connection. The `report_data` field of each attestation token carries the fingerprint of the SecretVM’s TLS certificate. This cryptographic binding mitigates impersonation and proxying attacks, ensuring the secure connection is tied directly to the attested hardware.
The Broader Trend of Trusted Inference Infrastructure
As the confidential computing market scales toward a projected $350 billion by 2032, the focus is shifting from basic isolation to verifiable trust. In confidential AI inference, the I/O subsystem and the execution environment must be part of a verifiable Trusted Computing Base (TCB).
“Confidential computing is only half the battle; verification is what enterprises actually buy,” said Alex Zaidelson, CEO of SCRT Labs. “By making Intel Trust Authority a standard feature of every Intel TDX-powered SecretVM, we’re making verifiable trust the baseline for the industry.”
For use cases ranging from confidential AI inference and financial services to healthcare and regulated compute, independent hardware attestation is no longer an optional security layer—it is the prerequisite for deployment. By embedding Intel Trust Authority into every SecretVM, SCRT Labs is establishing a new standard for verifiable trust in the AI infrastructure stack.
