Tenstorrent and Stealthium Partner to Secure AI's Critical 'Black Box'

SANTA CLARA, CA – July 30, 2026 – As artificial intelligence moves from research labs into the core of enterprise operations, a critical question of trust is emerging. Today, AI compute leader Tenstorrent and runtime security specialist Stealthium announced a partnership aimed squarely at this challenge, intending to bring unprecedented observability and security to the very hardware where AI workloads run.

The collaboration will integrate Stealthium's runtime observability platform directly with Tenstorrent's AI systems, which are built on an open RISC-V architecture. For customers in high-stakes sectors like finance, telecommunications, and energy, this partnership promises a new level of assurance, allowing them to monitor, verify, and secure the complex computations happening inside AI accelerators—an area that has, until now, largely operated as an opaque black box.

Addressing the 'GPU Blind Spot' in Enterprise AI

The explosive growth of AI has been powered by specialized processors, or accelerators, that can handle massive parallel computations. However, this has created what industry experts call a significant “GPU security gap.” Traditional cybersecurity and monitoring tools were designed for CPUs and general-purpose computing; they lack the visibility to inspect what happens during runtime inside an AI accelerator's dedicated memory and processing cores. This creates a “shared responsibility blind spot,” particularly in the cloud. While cloud providers secure their physical infrastructure, the customer is responsible for the workload running inside their allocated accelerator slice, often without the tools to do so effectively.

This isn't a theoretical problem. The risk is measurable and immediate. Financial services firms are running high-frequency trading and fraud detection models on hardware they cannot fully monitor. Healthcare organizations are processing sensitive patient data through AI systems with virtually zero execution-level visibility. An unprivileged process could potentially gain access to sensitive data being processed in a neighboring process on the same shared accelerator—a threat demonstrated by sophisticated attacks like 'GPUBreach'—and traditional endpoint security would be completely blind to it.

The partnership between Tenstorrent and Stealthium is designed to close this gap. By providing deep, hardware-level telemetry, the integrated solution will enable organizations to monitor for inefficient compute use, enforce workload isolation in multi-tenant environments, and, most critically, detect abnormal runtime behavior that could signal a security breach or operational failure before it cascades through the system.

A Look Under the Hood: How Stealthium Delivers Observability

Stealthium's core mission is to make accelerated compute observable, secure, and controlled. The company achieves this by moving beyond standard, high-level monitoring libraries. While tools like NVIDIA's Management Library (NVML) provide basic metrics, Stealthium has developed its own proprietary monitoring layer that interfaces more directly with the hardware drivers. This allows it to capture richer, more granular data with lower performance overhead, making it suitable for production-scale, real-time security.

The company's platform transforms this raw, low-level telemetry into what it calls “Hyperprints”—a high-level, actionable abstraction of GPU and AI workload activity. Instead of forcing security teams to parse cryptic kernel traces, Hyperprints provide a clear picture of how AI infrastructure is being used and can flag behavioral anomalies that indicate zero-day attacks or internal misuse. This capability to provide deep visibility across the entire software stack, from drivers to AI frameworks, is what sets its approach apart.

"AI is only as secure and trustworthy as the layer it runs on, yet for many that layer is currently invisible and indefensible," said Ahmed Shosha, CEO & Founder of Stealthium. "Stealthium exists to make AI accelerated compute observable, secure, and controlled. Tenstorrent is building exactly the kind of open, full-stack platform where that belongs from day one. Together we are offering customers an AI acceleration foundation they can see, verify, and trust as they scale."

Tenstorrent's Full-Stack Vision for Controllable AI

For Tenstorrent, this partnership is a significant step in its broader strategy to deliver more than just raw compute performance. Led by legendary chip architect Jim Keller—known for his work on Apple's A4/A5, AMD's Zen architecture, and Tesla's Full Self-Driving chip—Tenstorrent is building a comprehensive and open AI ecosystem. Its use of the RISC-V instruction set architecture is a deliberate move away from proprietary, closed systems, giving customers greater transparency and control over their hardware.

By building an open, full-stack platform, Tenstorrent enables ecosystem partners like Stealthium to integrate deeply and extend the platform's functionality. This strategy positions Tenstorrent not just as a chip vendor, but as a provider of enterprise-ready AI infrastructure that is transparent, flexible, and, with this partnership, demonstrably secure. It's a direct appeal to organizations looking to build private or sovereign AI clouds, where control and verifiability are non-negotiable.

"As enterprises and cloud providers move toward sovereign and private AI deployments, they need infrastructure that delivers high-performance compute and operational trust," said Amr Elashmawi, Vice President of Strategy & Business Development at Tenstorrent. "Tenstorrent’s open, full-stack AI platform gives customers greater control over their deployments, while Stealthium extends visibility and security at the runtime layer. Together, we can help customers build enterprise-grade AI infrastructure that is more secure, more transparent, and ready for production at scale."

The Bottom Line: From Sovereign Clouds to Secure Finance

Ultimately, the value of this partnership lies in its real-world business implications. For a financial institution, it means being able to prove to regulators that AI-driven trading algorithms are operating as intended and that customer data is isolated. For a national government building a sovereign AI capability, it provides the foundational trust needed to handle sensitive state data. For a cloud provider offering shared AI resources, it offers a powerful tool to guarantee workload isolation and bill customers more accurately based on actual resource consumption.

The integrated solution will soon be demonstrated on Tenstorrent’s customer cloud environment, a move that will shift the conversation from promise to practice. By embedding security and observability at the deepest layer of the AI stack, Tenstorrent and Stealthium are not just selling hardware or software; they are selling trust—a commodity that may prove to be the most valuable of all in the burgeoning AI economy.