EQTY Lab Announces Verifiable Runtime to Secure AI Agents Across the NVIDIA Enterprise AI Factory and NVIDIA OpenShell

• New zero-trust architecture for autonomous AI agents enables fine-grained runtime controls for multi-agent workflows with behavior-centric credentials and delegation chains with a root-of-trust in NVIDIA hardware.

• NVIDIA OpenShell demo at GTC showcases the first step in an agent governance roadmap that secures models and OpenClaw agents to power AI factories.

• EQTY Lab extends Verifiable Runtime on NVIDIA Confidential Computing (CC) to the NVIDIA BlueField Data Processing Unit (DPU) to create a silicon-based enforcement layer leveraging the NVIDIA DOCA Argus real-time situational awareness and threat detection.

SAN JOSE, CA --(BUSINESS WIRE)-- GTC 2026 – EQTY LAB announced today at NVIDIA GTC 2026 the launch of its new Verifiable Runtime, a solution engineered to secure and optimize autonomous AI agents running on the NVIDIA Enterprise AI Factory with NVIDIA OpenShell. The proliferation of AI agents from local desktop environments to the enterprise server core has created a massive trust gap, making verifiable behavior the final hurdle to scaling the agent economy. Verifiable Runtime addresses this challenge by leveraging EQTY’s Verifiable Compute Framework, delivering hardware-rooted oversight to ensure AI agents are secure and accountable.

Intelligent infrastructure to supervise agent deployments at scale.

On March 18, Founder and CEO Jonathan Dotan spoke at NVIDIA GTC to unveil a comprehensive roadmap to launch its Verifiable Runtime, which extends NVIDIA Confidential Computing (CC) to the NVIDIA BlueField Data Processing Unit (DPU).  Furthermore, by leveraging the NVIDIA DOCA Argus advanced threat detection framework, this integration creates a dedicated silicon-based enforcement layer. This full stack architecture allows the Verifiable Runtime to deliver hardware-attested execution that cryptographically proves three security properties:

1. Behavior alignment ensures agents are continuously audited at runtime, with trust rooted in the DOCA Argus Service, and the DPU enforcing policies managed in a separate control plane to uphold agent alignment throughout the agent lifecycle.

2. Delegation chains establishing who authorized each agent action, agent capability credentials, behavioral alignment confirming agents operate within defined guardrails, and untampered security postures protecting against runtime compromise.

3. Untampered security policies are protected from modification by the agent through native tools accessible only in the agent runtime and sensitive credentials that are not directly accessible by the agent.

“EQTY’s Verifiable Runtime gives enterprises the new governance layer that operates entirely outside the agent's reach.

Hardware-attested, cryptographically proven, and enforced at near-zero cost to the system, this new pairing of GPU and DPU is a game-changer for agentic productivity and security.”Yurko Jaremko, Head of Verifiable Compute, EQTY Lab

By combining NVIDIA Confidential Computing (CC) with NVIDIA BlueField DPU processors, EQTY Lab's Verifiable Runtime provides an unprecedented capability to deploy agents with a dedicated, silicon-based governance plane.

In the same way Kubernetes transformed containerized deployments into a predictable utility, Verifiable Runtime establishes a consistent orchestration layer for agent governance. This architecture unlocks an entire ecosystem of providers that can now leverage these standardized, hardware-rooted interfaces for real-time agent oversight and lifecycle management. For developers and providers, this means real-time oversight is no longer a custom engineering task—it is a native property of the infrastructure.

“As enterprises scale AI factories powered by autonomous agents, establishing verifiable trust, governance, and runtime security becomes essential to safely deploy multi-agentic systems across critical workflows.

Built on NVIDIA BlueField DPUs and the NVIDIA DOCA Argus framework, EQTY Lab’s Verifiable Runtime moves agent governance into a dedicated silicon enforcement layer that continuously observes, verifies, and protects AI agent activity.”Ofir Arkin, Sr. Distinguished Architect for Cybersecurity, NVIDIA

Cryptographically attested agent state and behavior traces provide a universal foundation for agent accountability, allowing any authorized entity to monitor and verify agent behavior without performance penalties.

Because the DPU has dedicated computing resources on ARM processors and direct memory access to the agent runtime on the host, governance enforcement can be implemented at near-zero cost to the system.

EQTY Lab’s own AI Gateway serves as the first reference implementation of this framework. The open nature of these verifiable traces ensures that any organization can build or integrate the oversight tools and trust infrastructure necessary to manage the burgeoning agent economy.

Verifiable Runtime on BlueField Benefits

• Physical separation from the host running the agent software, where the agent has no access and therefore cannot compromise the DPU’s detection and verification logic through native tools or any software-based attack.

• Cryptographic attestation generated by the DPU from hashes over code and execution state signed using keys on the DPU. These attestations are mathematically verifiable and cannot be forged by agent software.

• Dynamic enforcement incorporates deterministic policy checks and guardrail models that can be deployed to the DPU, close enough to inspect agent behavior in real-time while operating on an isolated environment with dedicated resources.

A Dedicated and Independent Layer of Trust

For enterprises navigating the shift to autonomous AI, Verifiable Runtime transforms agent governance from an operational liability into a competitive advantage. Traditional security perimeters stop where autonomous agent behavior begins. Verifiable Runtime brings zero-trust to the silicon layer, ensuring that all agent interactions are authenticated and attested in real time. This replaces vulnerable, static permissions with a hardware-rooted enforcement plane that tracks exactly what an agent is doing as it happens.

About EQTY Lab

EQTY Lab pioneers solutions that reinvent trust in AI. Our AI Integrity Suite applies cryptography to ensure that the governance of AI agents is verifiable, explainable, and accountable to all stakeholders. With applications in the public sector, life sciences, and finance, EQTY Lab is at the forefront of enabling AI innovation. To learn more about Verifiable Compute, go to vcomp.eqtylab.io and eqtylab.io.