By Nathan Simington, ESQ., Director of Policy & Strategy, Ericsson Federal Technologies Group
An Ericsson Federal Perspective on Operationalizing National Security AI
In the first week of June, two monumental presidential actions fundamentally reset the federal AI landscape.
On June 2, 2026, the White House issued Promoting Advanced Artificial Intelligence Innovation and Security, directing agencies to harden critical infrastructure against AI-enabled cyber threats, stand up a cybersecurity clearinghouse, and extend defensive tools to edge operators. Three days later, National Security Presidential Memorandum 11 (NSPM-11) commanded the defense and intelligence communities to accelerate AI adoption across warfighting missions, structured around four core pillars: Adoption, Adaptation, Assurance, and Accountability.
These are not aspirational guidelines; they are high-velocity mandates. With strict 30-day deadlines, they have already forced rapid operational shifts. CISA has issued its first binding operational directive (BOD 26-04), the Treasury-led clearinghouse is active, and the classified benchmarking framework for “covered frontier models” lands on August 1.
The national security enterprise is no longer debating whether to field AI—it is fielding it now, at scale. At Ericsson Federal, our analysis of these shifting mandates reveals a critical reality: this is an infrastructure problem long before it is a model problem.
Much of the public discourse centers on the software and algorithms themselves—how models are trained, governed, and tested. But at the tactical edge, the binding constraint on national security AI is not model quality. It is assured connectivity. The most capable model in the world delivers zero operational value if it cannot reach a distributed force, ingest sensor data in a contested environment, and close the decision loop inside a mission-critical window.
Adoption is a Reach Problem
NSPM-11’s Adoption pillar explicitly directs the federal enterprise to strip out barriers to rapid deployment. However, deploying AI in a national security context looks entirely different than deploying it in a commercial enterprise.
Mission environments require AI to function across highly distributed, mobile forces operating where commercial-grade connectivity is non-existent, denied, or actively compromised. Put simply: you cannot deploy AI to an edge you cannot reach.
Closing this gap requires modern, physical communications infrastructure:
- Private 5G and cellular networks that agencies wholly own and control.
- Multi-access edge computing (MEC) to position inference processing directly next to the sensor rather than a continent away.
- Non-terrestrial and satellite integration to guarantee resilience when terrestrial links are severed.
AI adoption at the pace the White House demands is fundamentally gated by the transport network that carries it.
Assurance is a Network Property
The Assurance pillar mandates that AI systems be robust, reliable, interoperable, and available. Crucially, these are not standalone attributes of an AI model; they are inherent properties of the network the model rides on. An accredited, perfectly optimized model sitting on a transport layer that collapses under load, fails to prioritize mission traffic, or cannot communicate across joint and coalition systems is not operationalized.
This is where the June directives converge. The Executive Order treats information systems as both an attack surface to be hardened and the exact channel through which AI defenses must be delivered. Hardening endpoints while leaving the transport layer soft is an incomplete defense.
True mission assurance requires secure-by-design radio access, zero-trust principles applied directly to transport, and network slicing to guarantee quality of service for mission-critical traffic. These capabilities are not adjacent to the assurance mandate—they are foundational to it.
Avoiding Lock-In Through Commercial Architecture
NSPM-11 is clear about the strategic risks of single-vendor dependence. At the network layer, mitigating this risk is entirely an architectural decision.
By mandates and necessity, agencies must favor open and standardized interfaces, multi-vendor interoperability, and modular architectures that allow components to be upgraded without re-platforming the entire environment. Leveraging the speed and scale of proven commercial technology development allows the government to innovate rapidly without inheriting legacy dependencies they cannot later unwind. This open principle should shape how agencies specify their network infrastructure, not just the software models they procure.
The Tactical Edge Case
To make this concrete, consider an edge intelligence workflow: distributed sensors feeding an AI model that must classify a signal and cue an immediate response in a contested electronic warfare environment.
If backhaul latency is unbounded, if the link degrades under electronic jamming, or if there is no resilient fallback path when the primary network drops, the model’s mathematical accuracy becomes entirely irrelevant because the decision loop fails to close in time.
The capability that saves the mission is not a better algorithm. It is a network explicitly engineered for that environment—combining local edge inference, slice-isolated transport, and transport-layer resilience. It is a connectivity problem with a connectivity solution.
The Capacity Question: Spectrum as the Raw Material
Everything required to field edge AI assumes the physical spectrum to build these private mission networks actually exists. Yet in the critical mid-band frequencies where coverage and capacity intersect, spectrum is increasingly scarce. This is the quiet constraint beneath the connectivity constraint—and current policy has placed a specific answer on the table: the lower 7 GHz band.
Under the One Big Beautiful Bill Act and a December 2025 presidential memorandum, the National Telecommunications and Information Administration (NTIA) is currently studying the 7.125–7.4 GHz band for future access. In 3GPP terms, this upper mid-band spectrum is close enough to classic mid-band to retain excellent propagation, while offering the wide, contiguous channels that high-throughput AI data workloads demand. It is the near-term raw material federal mission networks need most.
While a defense audience is naturally cautious—as the 7/8 GHz range currently hosts vital military satellite communications and radar—the path forward relies on coordinated, protected repurposing. The policy guardrails strictly require that any spectrum shifts do not materially impair national security missions. Done right, commercializing the lower 7 GHz does more than feed consumer 6G; it expands the globally standardized ecosystem that private tactical networks draw from, deepens the multi-vendor supply chain, and yields the licensed capacity required to operationalize AI at the edge.
The Path Forward: Where Ericsson Federal Fits
The organizations that successfully deploy AI will be those that reject silos and take a holistic view—one that treats applications, data, cybersecurity, and networks as a single, unified ecosystem.
This is the exact layer where Ericsson Federal operates. As a trusted, U.S.-based provider bridging commercial network innovation with rigorous government and defense missions, we focus on the connectivity foundation that the new mandates assume but do not build. We build the private cellular networks, edge computing deployment frameworks, and interoperable, zero-trust architectures necessary to move data securely where missions actually happen.
AI algorithms will continue to evolve on their own cycle. But whether the national security enterprise can actually field them—securely, rapidly, and at the tactical edge—depends entirely on the network underneath. The June directives set the ambition; meeting it starts one layer down.
Ultimately, AI needs a network—and mission success depends on getting that network right.
Â
