AI Etiquette: 10 AI Do’s and Do Nots for Government and National Security
Overview
When you use AI to assist your workflow, you are not handing over the pen. You are employing an automated scribe. If a document, a line of code, a procurement score, or an intelligence brief goes out under your credentials, you are the Author of the Record. Your cognitive sovereignty is your human responsibility to understand, validate, and defend that record. It is absolute and non-negotiable.
AI etiquette is the practice of interacting with artificial intelligence thoughtfully, respectfully, and responsibly, balancing efficiency with human judgment, transparency, and ethical and compliance considerations. This interactive scenario-based course exposes students to the key best practices for government use of AI.
Duration
1 day
Who Should Take This Course
Prerequisites
Participants should have experience interacting with LLMs. UMBC Training Centers’ AI Essentials and Prompt Engineering I courses or equivalent knowledge recommended.
Course Outline
AI Etiquette: 10 AI Do's and Do Nots for Government and National Security
- Module 1: The Principle of Cognitive Sovereignty
Objective: Understand the psychological dependency on automated systems and re-establish the absolute baseline of human command authority and legal accountability.
- The Paradigm Shift: Shifting the mindset from treating AI agents as an “all-knowing engine” to treating it as an automated, unverified scribe or entry-level assistant.
- The “Author of the Record” Standard: Establishing the legal, moral and ethical reality that if a line of code, procurement score, or intelligence brief goes out under human credentials, the human holds 100% accountability.
- The Fallacy of Speed: Distinguishing rapid text/code generation from actual project completion, conceptual mastery, or operational readiness.
Traceability: Directly establishes DO 1 (Treat AI like an entry-level assistant), DO 2 (Read, edit, and own every word), and DO 9 (Maintain personal domain expertise); directly counters the habits warned against in DON’T 1 (Blindly sending work), DON’T 3 (Treating AI as an accountable teammate), and DON’T 5 (The illusion of speed).
Module 2: The Operational Mandates
Objective: Equip operators with concrete, tactical boundaries to prevent data spillage, protect procurement integrity, and mitigate AI risk in high-stakes environments.
- Data Integrity & Verification: Mandatory line-editing and primary-source fact checking for all text, citations, and footnotes. Neutralizing the “Summary Trap” where AI strips out critical nuances and dissenting opinions.
- System Security, Environment Control & OPSEC: Preventing data spillage and internal data contamination. Zero tolerance for pasting sensitive data into unapproved public models and enforcing strict data-handling etiquette even within secure, siloed, or cleared government AI environments. Eradicating “Shadow AI” browser extensions and unvetted network tools.
- Procurement & Integrity: Safeguarding the procurement pipeline by enforcing strict compliance and ensuring AI usage does not compromise vendor proprietary data.
Traceability: Explicitly teaches DO 3 (Fact-check citations), DO 7 (Write objective prompts), and DO 10 (Verify data rights in procurement); actively enforces safeguards against DON’T 2 (Trusting AI
summaries), DON’T 7 (Mismanaging sensitive data across public or siloed models), DON’T 8 (Using Shadow AI), and DON’T 10 (Pasting proprietary bid data).
Module 3: Incident Analysis, Interactive Breakouts
Objective: Apply the core operational mandates to real-world government failure states, transforming
passive knowledge into defensive diagnostic skills. The Mission: Each room is assigned one specialized case study. They must identify the exact moment cognitive sovereignty was surrendered, map the failure to the corresponding Do’s and Don’ts matrix, and draft a corrective action plan to prevent systemic reoccurrence.
- Case A: Technical self-reliance during network blackouts or
- Case B: The risk of becoming a passive echo chamber for algorithmic hallucinations or
- Case C: How blind reliance on AI summaries triggers catastrophic bid protests or
- Case D: The danger of letting AI grade its own homework.
Traceability: Cultivates the real-world application of all do’s and do not’s. The cohort re-clusters in the main room. A representative from each delivers a 60-second operational takeaway, highlighting the universal truth across all fields: The moment human verification stops, systemic vulnerability begins.
Module 4: Technical Sovereignty & System Defense
Objective: Establish the systemic boundaries of human command authority over autonomous workflows and leverage adversarial testing to identify cognitive blind spots.
- Establishing Operational Execution Checks: Defining tight operational limits and human checkpoints before autonomous agents execute live actions, file deletions, or data transmissions.
- Adversarial Prompting: Teaching professionals how to actively test and intentionally “trick” their workflows to identify where analytical logic fails before deployment.
- Final Alignment: Reviewing the boundary lines of final human authority regarding citizen benefits, personnel reviews, and public funds allocation.
Traceability: Grounded in DO 5 (Applying the Principle of Least Privilege to agent boundaries) and DO 6 (Adversarial testing); establishes the hard operational limits demanded by DON’T
4 (Enforcing strict operational execution checks) and DON’T 9 (Handing over final resource/operational decisions to AI).