RE: 3.6m contract awarded5 Mar 2026 11:20
Thanks! I carried on with a bit of AI research, which is summarised below. Fundamentally this contract is likely part of a wider DARPA initiative called Red-C, all designed to be the backbone protection of the AI architecture all major players use to build their own AI infrastructure. It potentially has huge ongoing value to us I think (assuming we will own the IP as usual) , and brings prospect of follow on work.
https://www.darpa.mil/news/2025/red-c-watch-out-bus?utm_source=chatgpt.com
“DARPA recently awarded Narf Industries a contract titled “Helpful Strangers on a Bus: Protocol for Active Trust Probing and Recovery.” In cybersecurity terms, this refers to developing ways for hardware components inside a computer (GPUs, network cards, storage devices, accelerators, etc.) to actively verify and monitor each other’s behavior over internal system buses rather than assuming they are trusted. Modern systems rely on high-speed interconnects such as PCI Express and Compute Express Link that allow devices to communicate directly and access system memory. If a single device or its firmware were compromised—whether by malware, supply-chain tampering, or a rogue peripheral—it could potentially attack the rest of the system. The DARPA concept is to create protocols where devices can probe each other’s integrity, detect suspicious behavior, and help isolate or recover from compromised components.
The contract appears to fit within DARPA’s broader program Reclaiming Bus‑based Systems During Compromise (Red‑C), which aims to build self-monitoring and self-healing computer systems at the hardware-bus level. The idea is that devices connected to the same bus behave like “helpful strangers”: they don’t inherently trust each other but can collaboratively detect anomalies such as unusual memory access patterns, abnormal DMA activity, or inconsistent firmware responses. This approach is meant to address risks such as malicious PCIe devices, compromised firmware, or hardware implants that traditional operating-system security tools would never see because they operate below the OS layer.
The timing of this research is notable because the industry is rapidly moving toward CXL-based composable data centers and large AI compute systems, where CPUs, GPUs, memory pools, and accelerators from multiple vendors are dynamically connected across shared fabrics. Major players like NVIDIA, Intel, AMD, Google, and Microsoft are building infrastructure around these architectures. As the internal bus effectively becomes a high-speed network connecting dozens or hundreds of devices, security at the bus level becomes strategically important, particularly for AI and defense systems. If Narf is contributing protocols or detection techniques in this area, it places the company within a DARPA research stream that could plausibly lead to follow-on work as these architectures become more widespread.