News | Jan. 20, 2026
AI And The Reconfiguration Of Military Capability
The Department of War (DoW) is moving fast to integrate artificial intelligence across its workforce and mission sets. Under Secretary of War for Research and Engineering Emil Michael has called out historical under-use of AI and is backing enterprise adoption, from governance to deployment. The 2026 AI Strategy, plans to run commercial large-language models like Grok on classified and unclassified networks, and public announcements by Secretary of War Pete Hegseth at SpaceX all point to the same direction: AI is now part of how the institution operates.
This shift isn't just tech. It changes who has leverage in decision cycles, how information is filtered, and what options surface first. AI is becoming the interface through which operational knowledge is produced and acted on-affecting tactics, logistics, command, and strategic judgment.
AI Is Now A Decision Environment
Across DoW networks, AI systems decide which data matters, what patterns to elevate, and how to present courses of action. Decision advantage comes less from better sensors or faster munitions, and more from curating the space in which decisions happen. The FY 2026 National Defense Authorization Act (NDAA) echoes this by directing integration of commercial AI, governance frameworks, and cross-functional teams to manage and assess models.
AI is also treated as state capacity: an amplifier for military effectiveness, industry, and diplomacy. That means defense AI policy is tied to economic competitiveness and alliances, not just programs of record. The practical read for operations leaders: your data, your model lifecycle, and your command interfaces are now strategic resources.
What Operations Leaders Should Do Now
- Codify decision rights under speed. Define where AI informs, where it recommends, and where it may auto-execute within strict bounds. Preserve unity of command and decisive human judgment for use-of-force decisions.
- Stand up model operations. Create a model registry, version control, test plans, performance thresholds, drift monitoring, and ATO updates. Treat models like any other critical system with lifecycle discipline.
- Secure the decision environment. Enforce data lineage, integrity checks, and access control. Add adversarial testing, deception-resistance, and an incident response playbook for data poisoning or model manipulation.
- Integrate cleanly with command-and-control. For JADC2 and related constructs, define latency budgets, graceful-degradation modes, and manual fallbacks. No single point of failure, including the AI.
- Design for cognitive clarity. Make provenance, confidence, and underlying assumptions visible in the UI. Require dissent channels, override paths, and "two-key" release for high-consequence actions.
- Measure tempo and effects. Track decision latency, OODA compression, false positives/negatives, and escalation indicators. Balance speed with accuracy and mission risk.
- Train leaders in AI and ethics. Insert scenario-based drills, red teaming, and after-action reviews to build judgment under compressed timelines.
Competitive Pressure Without Blind Spots
Peer competitors are pursuing "intelligentized warfare"-AI embedded across domains, iterating in contact. That raises pressure to deploy fast to maintain relevance and deterrence. The risk: acceleration can crowd out debate, doctrine, and guardrails.
- Use stage gates that couple fielding pace with testing, safety cases, and escalation analysis.
- Make red teaming continuous: adversarial ML, deception, and exploitation scenarios.
- Run wargames that include cognitive effects: salience shifts, compressed timing, and cross-domain escalation.
- Keep procurement pathways open to commercial tools while enforcing evaluation and monitoring standards.
System-Level Risk You Must Own
AI introduces dependencies on data quality, model reliability, and the networks that carry both. Adversaries will target those layers because it's cheaper and often more impactful than kinetic action. Protecting data and models is protecting mission decisions.
- Set data contracts, gold datasets, and immutable logs for audits and forensics.
- Publish model cards, validate with third-party audits where possible, and require supply chain attestation for model and component sources.
- Exercise in contested conditions: degraded comms, spoofed feeds, corrupted data, and manipulated model outputs.
- Instrument drift detection and bias checks tied to defined thresholds and rollback triggers.
90-Day Implementation Plan
- Days 0-30: Inventory AI use by mission and system. Map decisions by consequence level. Assign accountable owners. Baseline latency, accuracy, and escalation metrics.
- Days 31-60: Draft doctrine updates for AI-mediated decision authority. Stand up a model registry and monitoring. Launch adversarial test team. Deliver a short, scenario-based training package for leaders.
- Days 61-90: Run a JADC2-integrated exercise with degraded-mode drills. Execute a tabletop on crisis escalation under AI-curated options. Review findings and adjust doctrine, interfaces, and safeguards.
References And Useful Links
- Joint All-Domain Command and Control (JADC2)
- NIST AI Risk Management Framework
- AI courses by job role - Complete AI Training
Bottom Line
AI now mediates how decisions are made, not just how data is stored or processed. If speed is going up, governance has to match it. Codify authority, keep accountability with command, protect data and models, and train leaders to read, question, and, when needed, override the machine-without losing tempo.
Your membership also unlocks:
