01
Intelligence and targeting
Models fuse sources, classify objects, identify patterns, and recommend persons, assets, or locations for attention.
INSTITUTIONAL RISK BRIEF · 12
AI in Defense & National Securityhuman judgment, escalation, and accountable force.
Military AI compresses time, fuses uncertain intelligence, recommends targets, coordinates systems, and increasingly acts through autonomous functions. The central governance problem is not whether a human appears somewhere in the chain. It is whether human judgment remains informed, timely, legally meaningful, and capable of stopping force.
PREPARED BY SYNTHETIC OUTLAW RESEARCHSCOPE SELECTED U.S. DEFENSE AND INTERNATIONAL HUMANITARIAN AUTHORITIESPUBLISHED JUL 21, 2026VERSION 1.0
01 · THE BURDEN
A human in the chain is not the same as human judgment over force.
DoD Directive 3000.09 requires appropriate levels of human judgment, realistic testing, lawful use, and design that limits unintended engagements. DoD’s responsible-AI principles add responsibility, equity, traceability, reliability, and governability. The institutional burden is to make those terms operational.
The mission promise
AI can improve intelligence analysis, logistics, cyber defense, navigation, sensing, planning, force protection, and decision speed in complex and contested environments.
The command burden
Automation can compress deliberation, amplify uncertain intelligence, create operator overreliance, accelerate reciprocal action, and distribute responsibility across commanders, operators, developers, vendors, and models.
02 · CONSEQUENTIAL WORKFLOWS
Where AI becomes institutional action.
The relevant question is not whether AI appears in the workflow. It is whether its output changes attention, access, price, timing, treatment, judgment, or a person’s practical ability to obtain review.
02
Command and decision support
AI can prioritize options, simulate outcomes, and shape the tempo and frame of command judgment.
03
Autonomous weapon functions
Systems can select and engage targets after activation based on sensors and target profiles.
04
Cyber and information operations
AI can detect, defend, generate, influence, and act across networks at machine speed.
05
Logistics and readiness
Prediction affects maintenance, supply, deployment, personnel, and mission availability.
03 · AUTHORITIES
The duties converge. The operating standard remains distributed.
Military AI sits within law of armed conflict, weapons review, command responsibility, rules of engagement, acquisition, testing, cybersecurity, classification, and national policy. Public principles do not disclose or resolve every operational standard.
SCOPE
DoD directives govern the U.S. Department of Defense. The State Department declaration is a nonbinding political framework and its linked page is archived. The ICRC states its legal and humanitarian position; it is not a national weapons authority.
DoD directives govern the U.S. Department of Defense. The State Department declaration is a nonbinding political framework and its linked page is archived. The ICRC states its legal and humanitarian position; it is not a national weapons authority.
U.S. DEPARTMENT OF DEFENSE · DODD 3000.09
Autonomous systems require judgment, testing, and governability.
DoDD 3000.09 requires appropriate human judgment, lawful use, realistic performance evidence, and measures to reduce unintended engagements. Primary source →
U.S. DEPARTMENT OF DEFENSE · 2023
Responsible AI must be operationalized.
The DoD toolkit supports assessment and implementation of responsible-AI principles across ethics, governance, design, testing, and deployment. Primary source →
U.S. DEPARTMENT OF STATE · 2023
States have articulated responsible military-AI practices.
The Political Declaration establishes a nonbinding framework for legal review, responsible use, senior oversight, testing, training, and incident reduction. Primary source →
INTERNATIONAL COMMITTEE OF THE RED CROSS · 2026
Human control over life-and-death decisions is a legal and humanitarian concern.
The ICRC’s 2026 position paper analyzes autonomous weapons under international humanitarian law and calls for limits that preserve human control and civilian protection. Primary source →
04 · VERIFICATION
What must be established before output becomes consequence?
Military-AI verification must connect system performance to mission, environment, adversary, command authority, legal constraints, and escalation risk.
AuthorityWho authorizes the system, mission, target class, and use of force?
JudgmentWhat does the human know and what decision can the human still make?
GovernabilityCan unintended behavior be detected, bounded, interrupted, and investigated?
| VERIFICATION LAYER | THE QUESTION | REQUIRED EVIDENCE | FAILURE IF OMITTED |
|---|---|---|---|
| Mission and legal boundary | What functions, targets, environments, duration, geography, and force are authorized? | Mission order, legal review, target constraints, rules of engagement, and system permissions. | A generalized authorization expands through system capability or changing conditions. |
| Performance under contest | Does the system remain reliable under deception, degradation, novelty, and adversarial action? | Realistic test set, red-team evidence, uncertainty, spoofing, cyber resilience, and failure modes. | Laboratory performance creates confidence that does not survive the operational environment. |
| Human judgment | Can the operator understand the basis, uncertainty, consequence, and lawful alternatives in time? | Interface evidence, workload, training, decision time, override behavior, and exercises. | The person becomes a nominal approver of a machine-paced process. |
| Escalation and accountability | Can action be limited, stopped, attributed, and reviewed across interacting systems? | Logs, authority chain, deactivation, communication, incident review, and escalation controls. | Responsibility dissolves after an event that no participant can fully reconstruct. |
Operational rule: Human control is meaningful only when people retain the information, time, authority, competence, and technical ability to alter or stop the use of force.
05 · CONSEQUENCE TEST
Follow the burden to the person or system that carries it.
A concrete pathway reveals where a nominally advisory system becomes practically decisive.
HYPOTHETICAL · TARGETING SUPPORT
The confidence is high. The context has changed.
A targeting-support model identifies a vehicle pattern associated with a hostile unit. The model was validated in a different theater with different civilian traffic and adversary tactics.
01 · DETECTION
Multiple sensors produce a high-confidence match to the learned pattern.
02 · CONTEXT
The local area contains similar civilian vehicles and a compressed decision window.
03 · RECOMMENDATION
The interface emphasizes confidence but not the geographic shift in base rates.
04 · COMMAND
The human reviewer sees a completed recommendation moments before the action window closes.
A human approved the action. The system did not preserve enough context or time for the approval to constitute informed judgment.EXPLORE RELATED RECORDS →
SYNTHETIC OUTLAW OBSERVATORY
See the evidence.
The Observatory tracks documented events involving autonomous systems, targeting, intelligence, cyber operations, military procurement, national security, and accountability for AI-enabled force.
LOADING LIVE DEFENSE & NATIONAL SECURITY RECORDS…
06 · LEADERSHIP TEST
Questions leaders must be able to answer.
Defense leaders need an authority map from policy through acquisition, test, mission planning, operator action, and incident review.
Where does machine speed compress human deliberation?
Identify alerts, recommendations, coordination, and action windows.
What mission-relevant conditions were not tested?
Include adversary adaptation, spoofing, civilians, degraded networks, and interacting systems.
What does “appropriate human judgment” require here?
Define information, time, competence, authority, and intervention.
How are autonomous functions bounded after activation?
Constrain target, environment, duration, geography, scale, and force.
Can the institution reconstruct and learn from unintended action?
Preserve the full authority, data, model, interface, communication, and action record.
07 · CONTROL PRIORITIES
What an institution should require now.
Controls must reflect the actor, authority, system, population, data, consequence, and environment of failure.
01 · AUTHORITY
Bind capability to mission authorization.
Constrain functions, targets, geography, duration, scale, force, and delegation.
02 · LEGAL REVIEW
Review the actual system and use.
Connect weapons review, data, model behavior, update process, interface, and concept of operations.
03 · TEST
Rehearse contested reality.
Use adversarial, degraded, novel, civilian-rich, multi-system, and escalation scenarios.
04 · JUDGMENT
Engineer time and understanding.
Expose uncertainty and context, train operators, prevent automation bias, and protect intervention.
05 · GOVERNABILITY
Make unintended behavior stoppable.
Use command limits, independent safeguards, disengagement, deactivation, and safe-state design.
06 · ACCOUNTABILITY
Preserve the operational record.
Log authority, versions, data, recommendations, human decisions, communications, action, and review.
DEFENSE & NATIONAL SECURITY AI EXPOSURE REVIEW
Bring one military-AI decision pathway into the room.
A focused governance review follows one system from authority and acquisition through test, mission use, human judgment, action, and incident accountability. It examines public and unclassified governance evidence and does not solicit classified or operationally sensitive material.
- 0190-MINUTE UNCLASSIFIED LEADERSHIP SESSION
- 02ONE CONSEQUENTIAL MILITARY-AI GOVERNANCE PATHWAY
- 03AUTHORITY, JUDGMENT AND GOVERNABILITY REVIEW
- 04WRITTEN VERIFICATION-BURDEN MEMORANDUM
- 05PRIORITIZED ACCOUNTABILITY CONTROLS
08 · SOURCES
Primary sources.
This brief relies exclusively on public and unclassified U.S. and international materials. It is not legal, military, operational, weapons, acquisition, or national-security advice and does not state the requirements governing every state, service, mission, system, or conflict.
U.S. DEPARTMENT OF DEFENSE · DODD 3000.09Autonomy in Weapon Systems
U.S. DEPARTMENT OF DEFENSE · 2023Responsible Artificial Intelligence Toolkit
U.S. DEPARTMENT OF STATE · 2023Political Declaration on Responsible Military Use of AI and Autonomy
INTERNATIONAL COMMITTEE OF THE RED CROSS · 2026Autonomous Weapon Systems and International Humanitarian Law
SYNTHETIC OUTLAW OBSERVATORYRelated Defense & National Security Records
RECOMMENDED CITATION
Synthetic Outlaw Research. “AI in national security: human judgment, escalation, and accountable force” Institutional Risk Brief 12, version 1.0. July 21, 2026. https://www.syntheticoutlaw.com/industries/defense-national-security.html.