Hub/Defense/Use Case 12
#12 of 15Tier 3 — Strategic Capability

Weapons System Targeting Support

AI provides decision support for targeting — identifying potential targets in imagery, assessing target validity against rules of engagement, estimating collateral damage, and generating targeting packages — with humans retaining all final release authority under LOAC and DoD AI ethics principles.

Latency Target
Seconds–minutes
Deployment
Classified On-Premises
Urgency Score
8 / 10
Maturity
Emerging
100%
Human Authority Retained for All Final Targeting Decisions

DoD Directive 3000.09 and the DoD AI Ethics Principles jointly require that human beings retain meaningful control over the application of lethal force. AI provides decision support — accelerating target identification, ROE analysis, and collateral damage estimation — but release authority rests with a human commander in all cases. This is a core design requirement that shapes the entire architecture of AI targeting support systems.

Key Context

Target Development
Hours → Minutes
AI processes imagery and multi-source data to develop targeting packages at a pace human analysts cannot sustain. Detection, characterization, and package assembly that took hours completes in minutes.
ROE Analysis
Seconds
LLM-powered ROE analysis applies current rules of engagement to proposed targets — flagging potential LOAC concerns, proportionality issues, and precautionary measure requirements for legal review.
Collateral Damage Estimation
Minutes
AI-assisted CDE modeling integrates target type, warhead selection, and surrounding population data — generating standardized CDE outputs for commander and legal review far faster than manual processes.

The Penalty Stakes

LOAC, DoDD 3000.09 & International Law Requirements
  • Laws of Armed Conflict (LOAC): All targeting decisions must comply with LOAC principles of distinction (discriminating between combatants and civilians), proportionality, and precaution. AI must support, not replace, this analysis.
  • DoDD 3000.09 — Autonomous Weapons: Weapons systems that select and engage targets without human action require SECDEF/Deputy SECDEF approval. AI targeting support maintains human release authority at all points — it is not an autonomous weapons system.
  • Immutable audit trail mandatory: Every AI recommendation, human decision, and action taken in the targeting process must be immutably logged. International humanitarian law investigations, congressional oversight, and LOAC compliance review all require complete, tamper-proof records.
  • JAG review integration: Judge Advocate General review is required for targeting decisions with complex LOAC implications. AI targeting support systems must produce outputs formatted for JAG review.

AI Performance vs. Rule-Based Systems

MetricRule-BasedAI-DrivenSource
Target identification in imageryObject detection + classificationAnalyst validates all potential targetsTS/SCI
Target type verificationMulti-source corroborationAnalyst confirms target identityTS/SCI
ROE compliance checkLLM against current ROELegal advisor reviews AI assessmentSECRET
Collateral damage estimationCDE modeling integrationCDM/JAG review requiredSECRET
Time-sensitive target workflowAutomated package generationApproval authority makes release decisionTS
Battle damage assessment (BDA)Post-strike imagery analysisAnalyst validates BDATS/SCI

Business Impact

AI vs. Manual Target ID Error Rate

Manual imagery analysis produces ~15–25% false identification rate. AI-assisted targeting (Maven program claims, corroborated by DoD exercise data) achieves ~5–8% false positive rate — a 3× improvement that reduces both missed targets and misidentified civilians.

Collateral Damage Estimation Accuracy

RAND 2022 analysis estimates AI-assisted CDE reduces collateral damage estimation error by 30–40% vs. manual methods — enabling more precise proportionality analysis and reducing both over- and under-estimation of civilian risk. Maven Smart System reduces analyst workload by ~80% for routine object detection tasks — freeing analysts to focus on complex cases, novel targets, and final validation rather than processing routine video feeds.

Infrastructure Requirements

Targeting data — target nominations, ROE, force disposition, CDE results — is among the most sensitive operational information. NEXUS OS hosts all targeting support inference within classified networks with immutable audit logging of every AI recommendation. NEXUS OS's LLM is configured against current Rules of Engagement — automatically flagging potential LOAC concerns in target nominations. JAG reviews and approves. Every AI targeting recommendation, with inputs and confidence scores, is immutably logged to a tamper-proof record. NEXUS Foundry trains target identification models on your specific target set — actual target types, appearances, and signatures relevant to the operational environment. NEXUS OS's architecture makes the human decision step explicit and mandatory, and compresses package assembly and LOAC analysis from hours to minutes — enabling commanders to act within TST windows while maintaining legal compliance.

Classified Infrastructure MandatoryROE-Compliant AnalysisImmutable Logging for Legal ReviewHigh-Confidence Target IDHuman Authority PreservedTime-Sensitive Target Support
AI Targeting Programs & Performance Data
Maven Smart System, JAGM ATR & the DoD Political Declaration
  • Maven Smart System (MSS) — NGA / CDAO / Palantir: $200M+ annual budget; operational across all CCMDs; 1,000+ FMV feeds simultaneously.
  • JAGM ATR (Joint Air-to-Ground Missile) — Lockheed Martin / Army: 92% classification accuracy; Full Operational Capability 2024; tri-mode seeker + AI ATR.
  • JDAM AI Seeker Enhancement — Boeing / USAF: 95%+ vehicle classification; AI/IR seeker variant — enables moving target engagement.
  • StormBreaker (SDB II) — Raytheon / USAF: Three-mode seeker; ATR + laser + radar; all-weather precision against moving targets.
  • DoD Political Declaration on AI Targeting — OSD (55 nations): March 2024; international principles for responsible AI in military targeting.