Blog

Safe, Edge, AI: The Software Properties Behind the AFA Themes

Written by Lynx | Sep 29, 2026, 1:00:01 PM

Part 1 pulled five themes out of AFA 2026: no sanctuary, unified C2, AI at machine speed, a high/low mix, and refresh under pressure. Strip away the domain language and each one reduces to a demand on the software underneath. Four properties do most of the work — and the real difficulty is that mission systems now need all four together, in the same box, at the same time.

 

Safe And Certifiable — Without Freezing Everything Else

Point defense, missile defense, and flight-critical functions carry the highest assurance requirements. Traditionally that meant isolating the safety-critical function on its own hardware and freezing it. But the same platforms now must host fast-changing software beside those functions. The property that matters is strong isolation with contained certification scope: safety-critical code proven to the required level, with a boundary firm enough that changes elsewhere do not force the safety case to be reopened.

 

Edge And Cloud-Independent

“No sanctuary” and contested communications mean a system cannot assume a link back to the cloud or the rear. Decisions must be made on the platform, in bounded time, whether connectivity holds. That calls for deterministic, real-time execution at the edge — predictable behavior under load, not just good average performance — and for autonomy that degrades gracefully when the network does.

 

AI And Accelerated Compute, Coexisting With Real-time

Machine-speed threat recognition means GPUs and AI accelerators moving into mission systems that also run hard real-time and safety-critical work. This is the sharpest technical challenge in the whole set. An AI workload sharing a processor with a fire-control or flight function cannot be allowed to disturb that function’s timing. Meeting it requires extending isolation and time-bounded execution across CPU and GPU together, so accelerated and safety-critical workloads coexist without interfering.

 

Any one of these properties is achievable alone. The difficulty — and the whole point — is holding all of them in a single system at once.

 

Open And Composable — So Many Functions Become One

Unified C2 and the high/low mix both require pulling many functions, often from many suppliers, into one integrated system. That is a modular-open-systems problem: the foundation must let independent functions share a platform through clean boundaries, so integration does not mean entanglement. Openness at the interface only works if the isolation underneath is strong enough to keep those functions from destabilizing one another.

 

Refreshable Across Decades

Because platforms last far longer than the technology inside them, every property above must survive repeated refresh. The measure of a good architecture is whether a new sensor, a new AI model, or an upgraded processor can be introduced by re-verifying only what changed — not the whole system. Upgradability is less a fifth property than the requirement that the first four endure over a decades-long life.

 

The Pattern

Every theme from the conference points at the same foundation: strong isolation, deterministic real-time behavior, safe coexistence of AI and safety-critical workloads across CPU and GPU, open integration, and the ability to refresh all of it over time. The final post in this series looks at how that foundation is built — and where it is already flying.

 

Contact Us Today