Verifiable Compute: Enhancing the Accountability of Confidential AI

Bringing Accountability to Confidential AI

Confidential AI sits at the intersection of two transformative technologies: artificial intelligence (AI) and confidential computing. While Confidential AI leverages secure computing environments to protect sensitive data, the question of how AI makes decisions remains a critical challenge. Organizations must trust not only that data is secure but also that AI systems are making decisions responsibly, explainable, and in compliance with regulatory standards.

This is where Verifiable Compute from EQTY Lab could play a transformative role in tandem with Intel Confidential Computing technologies. By combining confidential computing with verifiable records of AI workflows, Verifiable Compute provides organizations with the ability to audit and govern AI models throughout their lifecycle, ensuring they meet stringent accountability standards.

What is Verifiable Compute?

Verifiable Compute is a framework designed to make AI workflows explainable, accountable, and secure in real time. By building on Intel’s confidential computing technologies, EQTY Lab has developed a hardware-based solution that includes three major concepts:

Provable Records of Compliance:

Verifiable Compute generates immutable logs of AI decisions and processes, preserving critical artifacts for years to meet regulatory and audit requirements.

End-to-End Trust Zones:

Using Intel® Trust Domain Extensions (Intel® TDX) on 5th Gen Intel® Xeon® Processors, the framework extends confidential computing principles across the AI stack, including confidential VMs connected to NVIDIA’s H100/H200 GPUs and the upcoming Blackwell GPU architecture.

Independent Attestation:

The framework can integrate with Intel® Tiber™ Trust Authority, enabling independent runtime attestation to verify the integrity of AI environments.

These capabilities address critical gaps in today’s Confidential AI workflows by enhancing explainability, maintaining compliance, and ensuring the secure processing of sensitive data.

Supporting Ethical and Responsible AI

At Intel, we believe AI should be ethical, equitable, and trustworthy. Our Responsible AI philosophy emphasizes the importance of responsible practices across people, processes, systems, data, and algorithms. Verifiable Compute directly supports this vision by addressing two pressing challenges:

Explainability:

As AI decisions increasingly impact businesses and individuals, understanding how those decisions are made is vital. Verifiable Compute complements Confidential AI by enabling organizations to trace and interpret AI workflows without compromising sensitive data.

Compliance and Accountability:

In industries governed by stringent regulations like GDPR or HIPAA, maintaining provable compliance over time is non-negotiable. Verifiable Compute provides tools to help meet these requirements by securely preserving evidence of AI conformity and decision-making processes.

A Collaborative Effort

The Verifiable Compute framework exemplifies the power of collaboration in advancing secure and accountable AI. EQTY Lab’s innovative solutions, built on Intel’s cutting-edge technologies, bring another layer of trust to the confidential computing ecosystem. By integrating silicon-enabled security, independent attestation, and governance tools, Verifiable Compute sets a new standard for Confidential AI.

Looking Ahead

We at Intel are committed to pushing the boundaries of Confidential AI from edge to cloud. Tools like Verifiable Compute not only enhance security but also pave the way for more transparent and ethical AI systems. We are continuing our commitment and investment into the confidential computing ecosystem. As it continues to evolve, frameworks like Verifiable Compute will play a crucial role in bridging the gap between secure environments and responsible AI decision-making.

For more information, visit eqtylab.io/verifiablecompute and intel.com/confidentialcomputing

---

Anand Pashupathy is Vice President and General Manager of the Security Software & Services (S3) Division of Intel Product Assurance & Security (IPAS), where he leads a team of senior leaders to deliver security software technologies and services that activate and differentiate silicon security features and cultivate a high standard of security, operational, and cultural excellence. Anand is also responsible for Intel’s Confidential Computing vision, strategy, and execution.