AI-Powered Strategy for Uncertain National Security

A lot of national security teams are quietly accepting a bad bargain: either you write strategy that sounds confident but ages badly, or you stop trying and call it “agility.” Both options are losing moves.

The recent War on the Rocks conversation around whether strategy is even possible right now lands on an uncomfortable truth: modern statecraft is operating in a fog of shifting coalitions, domestic politics, gray-zone competition, rapid tech diffusion, and crises that jump theaters fast. Strategy isn’t dead—but our old methods for doing strategy are failing at the speed of events.

Here’s the stance I’ll defend: strategy is still possible, but only if it’s instrumented. Not “AI will write the strategy” instrumented. I mean strategy that’s continuously tested, stress‑checked, and updated using AI-driven scenario modeling, predictive analytics, and decision support tools—paired with humans who can set priorities and accept tradeoffs.

Why strategy feels impossible right now

Strategy feels impossible because the planning cycle is slower than the threat cycle. Many national security processes still assume you can set aims, allocate means, and execute over years with only minor course corrections. That assumption breaks when adversaries iterate weekly, supply chains wobble monthly, and domestic constraints can flip within an election cycle.

The War on the Rocks discussion highlights the real-world friction: U.S. statecraft isn’t operating in one tidy “grand strategy” lane. It’s juggling Latin America, Europe, the Middle East, and the Indo-Pacific—each with different escalation dynamics, alliance politics, and industrial constraints. This isn’t just complexity; it’s coupled complexity. A decision in one theater changes risk in another.

The modern strategy tax: uncertainty, speed, and coupling

Three factors make today’s environment particularly punishing:

1. Uncertainty is structural. Opponents hide intent, employ proxies, and blend military and non-military pressure.
2. Speed is asymmetric. A small actor can force tempo with drones, cyberattacks, or information operations.
3. Coupling is unavoidable. Munitions, ISR assets, ship maintenance, satellite coverage, and political attention are shared across theaters.

If your strategy process can’t represent these interactions, you’re not “doing strategy”—you’re producing talking points.

Strategy vs. vibes

A useful one-liner for teams that need to reset expectations:

If your strategy can’t name what you will stop doing, it’s not strategy—it’s vibes.

That’s where many national security plans break down: not on lofty goals, but on the hard tradeoffs about time, readiness, and finite resources.

What AI can actually do for strategy (and what it can’t)

AI doesn’t replace strategy; it restores the feedback loops strategy requires. The key value is not “prediction” in the fortune-teller sense. It’s faster sensemaking, better hypothesis testing, and earlier detection of second-order effects.

Used well, AI can help answer questions that strategy documents often dodge:

• What indicators would tell us our assumptions are failing?
• Which actions raise the probability of escalation, and where?
• If we surge forces here, what breaks elsewhere?
• What’s the fastest path for an adversary to impose costs on us?

AI-enabled scenario modeling: turning assumptions into testable claims

Traditional tabletop exercises are valuable, but they’re episodic and labor-intensive. AI-driven scenario modeling makes scenario generation cheaper and more continuous.

Practical ways teams are using it:

• Branch-and-sequel generation: Create plausible “next moves” for multiple actors and identify decision points.
• Monte Carlo simulations: Explore distributions of outcomes rather than single narrative paths.
• Adversary course-of-action modeling: Stress-test your plan against adaptive opponents.

The win is simple: fewer surprises from assumptions you forgot you made.

Predictive analytics: early warning that isn’t just “more alerts”

Most organizations don’t have an early warning problem—they have a signal integration problem. Predictive analytics can fuse disparate indicators into decision-relevant warnings.

Examples (kept at a conceptual level):

• Logistics disruptions + procurement anomalies + online recruiting patterns → potential mobilization preparation
• Coordinated cyber probing + disinformation seeding + diplomatic messaging shifts → pre-crisis shaping activity
• Maritime AIS irregularities + satellite change detection + insurance pricing shifts → sanctions evasion routes evolving

The output shouldn’t be “high risk” dashboards. It should be specific triggers tied to pre-approved options.

Autonomous systems: the strategic value is tempo and persistence

In many theaters, the biggest advantage isn’t exquisite capability—it’s presence and persistence at scale. Autonomous systems (air, surface, subsurface, and cyber) can create strategic options by:

• Extending ISR coverage without burning scarce crewed platforms
• Increasing maritime domain awareness and counter‑drone capacity
• Compressing decision cycles by automating routine detection-to-tasking workflows

But autonomy also raises real issues: escalation risk, rules of engagement complexity, and brittle behavior under adversarial deception. Which brings us to the part too many AI pitches skip.

What AI can’t do: set national priorities or absorb moral risk

AI can rank options, flag anomalies, and estimate outcomes. It can’t legitimately decide:

• whose security matters more when interests collide
• how much risk to accept for deterrence
• what level of collateral damage is politically and ethically acceptable

Those are human judgments. AI is the instrument panel; leadership still flies the aircraft.

A practical model: “instrumented strategy” for defense teams

Instrumented strategy means your strategy has sensors, thresholds, and update mechanisms—like an operational system. It’s written to be measured.

Here’s a workable blueprint I’ve seen succeed in complex organizations.

1) Start with a strategy that’s narrow enough to execute

If your strategy has 12 priorities, it has none. A strong starting point is:

• 1–2 primary objectives
• 2–4 supporting objectives
• a short list of things you will not do

This is especially relevant when U.S. policy must span multiple regions. The War on the Rocks panel’s “spicy takes” across Latin America, Europe, the Middle East, and the Indo-Pacific point to the same constraint: you can’t surge everywhere indefinitely.

2) Translate assumptions into “tripwires” and measurable indicators

Every strategy rests on assumptions about allies, adversaries, industry, and domestic support. Instrument them.

Create:

• Assumption statements (clear and falsifiable)
• Indicators (what would change our view?)
• Tripwire thresholds (when do we act?)
• Pre-baked options (what do we do if it triggers?)

AI helps most on the indicator and threshold layers: collecting, fusing, and highlighting change.

3) Build a theater-coupling view of resources

The fastest way to fail strategically is to manage each theater as if it’s independent. Defense leaders need a cross-theater coupling map of:

• munitions stocks and replenishment timelines
• ISR coverage and competing demands
• maintenance backlogs and surge limits
• cyber defense capacity and incident response bandwidth

AI-enabled planning tools can model these constraints and show the hidden tradeoffs before you commit to them.

4) Run “continuous red teaming” with AI and humans together

The goal is not to “beat the red team.” The goal is to expose fragility.

A strong cadence looks like:

• Weekly: AI-supported anomaly and indicator review
• Monthly: short scenario sprints (3–5 plausible branches)
• Quarterly: deep red-team exercise on the most stressed assumption

AI can generate branches and identify patterns. Humans decide which branches are credible, which are dangerous, and which are politically unrealistic.

Where AI helps most right now: three national security use cases

The best near-term ROI comes from AI systems that reduce decision latency and improve resource allocation—without requiring perfect prediction.

Use case 1: Mission planning and course-of-action comparison

AI-assisted mission planning can:

• compare multiple courses of action against objectives and constraints
• surface logistics and sustainment risks earlier
• highlight “unknowns” that matter (e.g., gaps in ISR coverage)

This supports the core strategic question: can we do what we’re claiming we can do?

Use case 2: Dynamic threat assessment in intelligence and cyber

AI can fuse telemetry and open-source signals into a dynamic threat picture that updates as conditions change. The strategic payoff is not omniscience—it’s fewer blind spots during transition moments, when adversaries often act.

If you’re serious about AI in defense and national security, prioritize:

• data lineage and provenance
• model evaluation under adversarial conditions
• workflows that connect warnings to decisions

Use case 3: Autonomous ISR and counter‑UAS for persistent presence

Autonomous systems are often framed as tactical tools. Their strategic value is that they change the cost curve of presence.

Persistent sensing and rapid cueing can:

• raise the cost of gray-zone actions
• reduce surprise and shorten response timelines
• free high-end assets for deterrence tasks

The caution: autonomous systems can also create escalation risk if commanders don’t have clear control policies and audit trails.

People also ask: Can AI “save” strategy?

AI can’t save a strategy that avoids tradeoffs, but it can keep a serious strategy from drifting into fantasy.

If you use AI to:

• test assumptions continuously
• quantify resource tradeoffs across theaters
• detect inflection points early
• rehearse options before crises

…then strategy becomes more durable under uncertainty.

If you use AI to generate polished narratives without decision authority, validated data, and accountability, you’ll get prettier documents—and worse outcomes.

How to evaluate an AI strategy tool before you buy it

Procurement and adoption fail when teams buy demos instead of decision advantage. Here’s a blunt checklist.

• Decision clarity: Which decision does it improve (and by how much time or accuracy)?
• Data reality: Can it work with your messy, incomplete data today?
• Adversarial robustness: How does it behave under deception and poisoned inputs?
• Auditability: Can you trace why it recommended something?
• Integration: Does it fit existing planning and intel workflows, or replace them?
• Governance: Who owns model updates, thresholds, and escalation policies?

One sentence worth repeating internally:

If you can’t audit it, you can’t operationalize it.

Strategy is possible—if you treat it like a living system

The War on the Rocks conversation circles a frustration many practitioners feel: the world is messy, leadership changes, crises cascade, and “strategy” can start to sound like a luxury item.

I see it differently. Strategy is still the only way to make scarce resources mean something. But in 2025, strategy has to be continuously measured and updated—especially across Latin America, Europe, the Middle East, and the Indo-Pacific, where actions in one theater can rapidly create liabilities in another.

If you’re building capabilities in the AI in Defense & National Security space, the bar is clear: help leaders make tradeoffs faster, with fewer blind spots, and with stronger accountability. Strategy doesn’t need to be abandoned. It needs to be instrumented.

Where are you still relying on quarterly briefs and static assumptions for decisions that change weekly—and what would it take to put real sensors on that part of your strategy?