7.8.2 — Military AI Systems — maturity: live
AI-Assisted Mission Planning
Using satellite-fed AI to generate, evaluate and refine military mission plans in near-real-time against live threat, weather and logistics data.
When every mission variable — weather windows, asset availability, threat envelopes, rules of engagement — must converge in minutes, AI running on sovereign space infrastructure is the only planner fast enough to be trusted.
Mission planning has always been a race against time: commanders must synthesise terrain, weather, threat positions, asset availability and rules of engagement into a coherent scheme of manoeuvre, often in hours or minutes. Traditional planning tools are static and analyst-heavy; they cannot ingest a continuous stream of satellite imagery, signals intelligence and atmospheric data and reprice options dynamically as conditions change. The result is plans that are stale before they are executed, and decision-makers who are always catching up.
A sovereign satellite constellation changes the input quality and the tempo simultaneously. Wide-area optical and SAR passes every 30-90 minutes provide current ground truth on threat disposition; HF/VHF RF survey payloads flag emitter changes that indicate force movements; weather sounders feed atmospheric models used for routing and effects prediction. An AI planning engine ingests all of this through a hardened, sovereign data pipeline, evaluates thousands of course-of-action permutations against doctrine-encoded objectives, and surfaces the top candidates with confidence scores, risk trade-offs and logistics dependencies attached.
The operational outcome is a commander who can compress the planning cycle from six hours to under sixty minutes and re-plan mid-mission when satellite cues indicate the situation has changed. Critically, the AI layer is trained on national doctrine, national order-of-battle data and national threat libraries — not a vendor's sanitised training set. That distinction is the difference between a tool that reflects your strategy and one that reflects someone else's.
Frequently asked
What exactly does an AI-assisted mission planner do that a human staff officer cannot?
It simultaneously evaluates thousands of course-of-action (COA) permutations — integrating real-time satellite imagery, SIGINT, weather forecasts, logistics states, and rules-of-engagement constraints — in under two minutes. A human staff team doing the same work typically needs 4–12 hours. The AI does not replace the commander's judgement; it collapses the option space so the commander can focus on the decision rather than the data.
Why does the satellite layer matter — couldn't this run on ground-based intelligence alone?
Ground-based sources (HUMINT, fixed radar, liaison reporting) are slow, geographically patchy, and often withheld by allies under caveats. A sovereign LEO constellation gives the planner a persistent, unmediated, all-weather picture of the battlespace updated every 30–90 minutes. Without that refresh rate, the AI is reasoning on a map that is already wrong.
Is this capability already operational, or is it still experimental?
Several allied nations have live deployments. The US JADC2 initiative integrates AI planning tools across services; the UK's Project MORPHEUS demonstrated AI-assisted joint planning in exercises by 2023; Israel's Unit 8200 has used algorithmic targeting tools in active operations. The maturity tag on this page ('live') reflects real operational deployments, though depth and integration vary significantly by nation.
How do we avoid the AI recommending plans that violate international humanitarian law?
Responsible deployment requires hard-coded constraint layers — essentially a rules-of-engagement engine that filters any COA touching protected sites, civilian density thresholds, or prohibited weapons effects before the plan is surfaced to the commander. IEEE 7000-2021 and the ICRC's human-control framework both provide design guidance. The constraint layer must be sovereign-controlled and auditable; outsourcing it to a vendor is legally untenable.
What is the difference between AI-assisted mission planning and autonomous targeting?
Mission planning AI recommends sequences of actions — who moves where, when, using which assets — with a human approving the plan. Autonomous targeting AI selects and engages individual targets without per-engagement human authorisation. The former is widely deployed; the latter is the subject of active UN GGE negotiations on Lethal Autonomous Weapons Systems (LAWS). Keeping the distinction clear in system architecture protects nations from treaty liability.
How many satellites does a nation actually need to feed a meaningful AI planning system?
A useful baseline is 12–16 LEO EO/SAR microsatellites for a regional theatre, yielding roughly 90-minute revisit. For global coverage with sub-60-minute revisit — sufficient for high-tempo operations — the figure rises to 40–60 satellites. Supplementing with commercial data licences from Planet or ICEYE can bridge gaps during initial constellation build-out, but every commercially sourced layer is a sovereignty dependency that must be quantified.
Can a smaller or middle-income nation realistically build this capability, or is it only for major powers?
The cost of entry has fallen sharply. A six-to-twelve satellite nanosatellite constellation with onboard processing now costs $80M–$200M to build and launch, versus $1B+ a decade ago. Open-source AI frameworks (PyTorch, TensorFlow) mean the modelling layer is not inherently proprietary. The genuine barriers are trained personnel, secure ground infrastructure, and the classified datasets needed to tune models — all of which require sustained political commitment, not unlimited budgets.
What happens to the AI planner if communications are jammed or satellites are degraded?
A well-designed sovereign system pre-computes contingency COA libraries during periods of connectivity and caches them at the operational edge — ships, forward bases, aircraft — so the planner can continue reasoning offline. This requires deliberate resilience engineering and regular exercises to validate degraded-mode performance. Nations relying on a commercial cloud-hosted AI service have no equivalent fallback when the link is cut.