AI Ethical Use

Start Here If You Are New to AI Ethical Use

ARTIFICIAL INTELLIGENCE in the Marine Corps must be used ETHICALLY and RESPONSIBLY. The DoD Responsible AI Strategy (June 2022) establishes the framework. MCO 5231.4 requires AI capabilities to be RESPONSIBLE, EQUITABLE, TRACEABLE, RELIABLE, and GOVERNABLE. MARADMIN 635/24 (signed 30 December 2024, supersedes MARADMIN 602/24) directs COMMSTRAT-specific guiding principles. Six guiding principles for COMMSTRAT use of AI. Use of AI must align with DOD Principles of Information (truthful, accurate). Only DOD-approved AI technologies. NO photo-realistic imagery, video, or news stories for public dissemination using AI. AI annotation required in caption and metadata. Human oversight required. Protect against atrophy of creative and technical skills. While written for COMMSTRAT, these principles inform broader Marine Corps AI use. Per MCO 5231.4 and MARADMIN 635/24.

This page walks you through AI ethical use from your seat as a Marine. By the end you will know.

• The DoD Responsible AI Strategy
• MCO 5231.4 five attributes of responsible AI
• COMMSTRAT-specific principles per MARADMIN 635/24
• The supersession of MARADMIN 602/24 by 635/24
• DOD Principles of Information alignment
• Human oversight requirements
• AI annotation in caption and metadata
• Imagery and video restrictions
• Limitations of GenAI (hallucination, bias)
• Common ethical pitfalls

Every fact on this page comes from MARADMIN 635/24, MCO 5231.4, DoD Responsible AI Strategy, DEPSECDEF Memorandum (26 May 2021), Executive Order 14110, OMB Memorandum M-24-10, and DoDI 5040.02 (DoD Visual Information Style Guide).

The DoD Responsible AI Strategy

The DoD Responsible AI Strategy and Implementation Pathway (June 2022) establishes the framework for ethical AI in the Department.

Six Tenets

Per MCO 5231.4 paragraph 1, the Marine Corps adheres to six tenets for responsible AI use and implementation. These tenets align with the DoD Strategy and DEPSECDEF Memorandum (26 May 2021).

Five Attributes of Responsible AI

Per MCO 5231.4. AI capabilities must be.

• RESPONSIBLE. Marine Corps personnel exercise appropriate levels of judgment and care while remaining responsible for the development, deployment, and use of AI capabilities.
• EQUITABLE. The Marine Corps takes deliberate steps to minimize unintended bias in AI capabilities.
• TRACEABLE. AI capabilities are developed and deployed such that relevant personnel possess an appropriate understanding of technology, development processes, and operational methods.
• RELIABLE. AI capabilities have explicit, well-defined uses, and their safety, security, and effectiveness are subject to testing and assurance within those defined uses.
• GOVERNABLE. The Marine Corps designs and engineers AI capabilities to fulfill intended functions while possessing the ability to detect and avoid unintended consequences and to disengage or deactivate deployed systems that demonstrate unintended behavior.

COMMSTRAT-Specific Principles Per MARADMIN 635/24

MARADMIN 635/24 directs the Communication Strategy and Operations (COMMSTRAT) community to adhere to six guiding principles. The Communication Directorate (CD) released the MARADMIN. While addressed to COMMSTRAT, the principles inform broader Marine Corps AI use.

Principle 1. Alignment with DOD Principles of Information

Use of AI will align with DOD Principles of Information to ensure information released is TRUTHFUL and ACCURATE.

• AI can be used to increase editorial efficiency.
• AI can be used to help inform understanding.
• AI can support legitimate command narratives.
• AI can counter mis-, dis-, and malinformation.

Principle 2. Protection of Information

To uphold the responsibility to protect information when utilizing AI tools, ensure adherence to.

• SECURITY standards.
• ACCURACY standards.
• PRIVACY standards.
• PROPRIETY standards.

ONLY DOD-APPROVED AI TECHNOLOGIES will be used.

Principle 3. Imagery and Video Restrictions

Imagery and video are uniquely truthful and compelling mediums for informing understanding. As such.

• The Marine Corps will NOT employ AI to create photo-realistic imagery, video, or news stories for public dissemination.
• AI will only be used to REVIEW WRITTEN PRODUCTS.
• AI will only be used to ASSIST WITH CORRECTIVE TECHNIQUES to visual information as specified in DoDI 5040.02 (DoD Visual Information Style Guide).
• Products adjusted with AI MUST ANNOTATE that adjustment in both CAPTION and METADATA (e.g., basic correction of color done with AI).

Principle 4. Automation and Information Environment Understanding

AI can be used to.

• Automate processes.
• Enhance understanding of the information environment.

Principle 5. Protection of Trust

It is the responsibility of COMMSTRAT to protect the trust of key publics and the integrity and legitimacy of commands and missions through the content released.

Principle 6. Skill Maintenance

Over-reliance on AI can lead to ATROPHY of creative and technical skills and challenge proficiency in conventional content creation methods.

• Build and maintain proficiency in conventional content creation methods.
• Use AI to augment, not replace, human skill.

Supersession of MARADMIN 602/24

MARADMIN 602/24 (signed 13 December 2024) was the initial COMMSTRAT AI ethics MARADMIN. It was SUPERSEDED by MARADMIN 635/24 (signed 30 December 2024). MARADMIN 602/24 is now CANCELLED.

Key Differences in 635/24

• References MCO 5231.4 (added).
• References OMB Memorandum M-24-10 (added).
• Refined language on AI as an editorial efficiency tool.
• Refined language on countering mis-, dis-, and malinformation.
• Refined language on imagery and video annotations.

For current guidance, use MARADMIN 635/24 only.

DOD Principles of Information Alignment

The DOD Principles of Information establish that information released by the DoD must be.

• Truthful.
• Accurate.
• Timely.
• Responsive to the public's right to know.

AI cannot be used to deceive, manipulate, or spread misinformation. AI must be a tool for understanding and efficiency, not a tool for deception.

Human Oversight Requirements

Across all AI principles, HUMAN OVERSIGHT is a constant theme.

Key Aspects

• Human oversight in the release of information.
• Critical evaluation of AI outputs before use.
• User accountability for products derived from AI.
• Recognition of AI limitations.

Why Human Oversight Matters

AI tools are non-deterministic. They may produce.

• Inaccurate information.
• Biased outputs.
• Incomplete responses.
• Fabricated references (hallucinations).

Humans remain the final authority on what is released and used.

AI Annotation Requirements

Per MARADMIN 635/24 Principle 3, products adjusted with AI MUST be annotated in both.

Caption

A visible annotation in the caption explaining the AI adjustment.

Example. "Basic correction of color done with AI."

Metadata

A non-visible annotation in the file metadata explaining the AI adjustment.

When Annotation Is Required

• Color correction with AI.
• Cropping or resizing with AI assistance (when in scope of DoDI 5040.02).
• Other corrective techniques per DoDI 5040.02.

When Annotation Is Not Required

• AI used only to review written products (no annotation on the written product itself).
• AI used to understand the information environment (analysis output not for public release).

Imagery and Video Restrictions

Per MARADMIN 635/24 Principle 3.

Prohibited

• AI-generated photo-realistic imagery for public dissemination.
• AI-generated video for public dissemination.
• AI-generated news stories for public dissemination.

Permitted

• AI review of written products.
• AI-assisted corrective techniques (color correction, etc.) within DoDI 5040.02 scope, with annotation.
• AI analysis of imagery for internal understanding.

Why These Restrictions

Imagery and video are uniquely truthful mediums for conveying actual events. Marine Corps credibility depends on the truthfulness of visual information.

Limitations of GenAI

Hallucination

GenAI fabricates facts, citations, and references with confident-sounding language. Always verify.

Bias

GenAI training data contains biases. Output may reflect those biases.

Recency

GenAI training data has a cutoff date. Recent information may be missing or wrong.

Misunderstanding Context

GenAI may misinterpret the user's intent or context.

Inability to Verify Sources

Even with web grounding, GenAI may misattribute information.

Common Ethical Pitfalls

Treating AI Output as Authoritative

Marines copy AI output into official documents without verification. Errors propagate.

Mitigation. Verify every fact, citation, and figure.

Using Unapproved AI Tools

Marines use commercial AI tools for official work involving CUI or sensitive information.

Mitigation. Use only DoD-approved AI (GenAI.mil, CamoGPT). Personal use of commercial AI is permissible for personal matters only, not Marine Corps work.

Failing to Annotate AI-Assisted Imagery

Marines use AI to adjust photos without annotation, violating MARADMIN 635/24.

Mitigation. Annotate AI adjustments in caption and metadata.

Generating Photo-Realistic Content for Public Use

Marines create AI imagery for public release, violating Principle 3.

Mitigation. Use AI only for written review and authorized corrective techniques.

Skill Atrophy

Marines depend on AI to the point that conventional skills atrophy. When AI is unavailable, work suffers.

Mitigation. Practice conventional methods. Use AI to augment, not replace.

Over-Sharing With AI

Marines paste PII, PHI, or classified information into GenAI tools.

Mitigation. Strictly limit AI input to CUI or below.

Cross References

• For overall AI framework, see AI Overview and Authorities
• For the GenAI.mil platform with approved use, see GenAI.mil Enterprise Platform
• For other approved AI tools, see CamoGPT and Other Approved AI Tools
• For governance and approval authority, see AI Governance and Approval
• For day-to-day ethical practice, see Practical AI Use for Marines
• For training that covers ethics, see Basic AI Course (Mandatory)

Quick Reference

• Authority. MCO 5231.4 and MARADMIN 635/24.
• Five attributes. Responsible, equitable, traceable, reliable, governable.
• Six COMMSTRAT principles. DOD Principles of Information, info protection, no AI imagery/video, automation, trust protection, skill maintenance.
• Annotation. Caption AND metadata for AI-adjusted products.
• Prohibited. AI photo-realistic imagery/video/news for public dissemination.
• Verification. Always verify AI output before use.
• Tools. Only DOD-approved.
• Superseded. MARADMIN 602/24 is cancelled, 635/24 is current.

Records To Keep

• Awareness of MARADMIN 635/24 (read it directly)
• Awareness of MCO 5231.4 responsible AI principles
• Documentation of any AI annotations on products you produce
• Notes on AI tools used (for accountability if work is questioned)

If you have a question about AI ethical use, your unit S-1 routes to DC I or COMMSTRAT for COMMSTRAT-specific questions. For general ethical guidance, your chain of command and Staff Judge Advocate (SJA) are resources.