MIT Technology Review publishes June 2026 eBook compiling six military AI decision stories from April 2025-April 2026

The collection documents AI systems shifting from data analysis tools to direct recommendation generators for targeting, logistics, and rules-of-engagement choices across U.S., allied, and adversary forces. Stories cite internal program documents showing model outputs routed into joint all-domain command systems with human sign-off windows reduced to seconds in simulated exercises. No public release of training data, decision logs, or failure-mode test results accompanies the reported deployments.

DoD budget justification books for FY2025 and FY2026 list $1.8 billion in AI-enabled decision programs, while a 2024 DARPA Explainable AI follow-on report records 37 percent of evaluated models producing untraceable recommendation chains under contested conditions. Parallel NATO Science and Technology Office working papers from 2025 note absence of standardized audit interfaces across member states, creating verification gaps when models cross national boundaries.

Operational integration therefore bypasses existing public oversight mechanisms designed for weapons rather than advisory layers. Commanders receive model outputs inside existing C2 interfaces, rendering the distinction between tool and decision participant invisible to external review. This pattern repeats earlier opaque sensor-fusion rollouts that later required retroactive policy patches after field incidents.

Next milestones include scheduled FY2027 exercises embedding AI recommendations at corps level with automated escalation triggers and a pending Defense Innovation Board update on accountability frameworks scheduled for Q4 2026.