State, Local, Tribal, and Territorial (SLTT) emergency management agencies are being asked to do more with less than at any point in the profession’s modern history. In a 2025 Argonne National Laboratory survey of 1,689 local emergency management agencies, more than half reported having one or no permanent full-time employees. State directors surveyed by Deloitte and the National Emergency Management Association named budget constraints and a competitive job market (81% each) and a shortage of qualified candidates (64%) as their top workforce challenges. At the same time, the planning workload keeps growing: emergency operations plans, hazard mitigation plans, THIRA/SPR submissions, grant justifications, and after-action reports, each carrying its own compliance requirements and cross-references.
Into this gap arrives generative artificial intelligence, and with it a common anxiety inside the profession: that AI will automate away the role of the planner.
This paper challenges that assumption. Emergency management is fundamentally a human endeavor rooted in relationship management, community trust, and multi-agency coordination. No model can sit across the table from a hospital administrator and negotiate a resource-sharing agreement. What AI can do, when properly governed, is absorb the administrative weight that currently keeps planners at their keyboards instead of in their communities.
The Capacity Problem Is Already Here
The emergency management workforce conversation usually centers on response staffing. The quieter crisis is in preparedness. Local programs run by a single coordinator, or by an official wearing emergency management as a second or third hat, are the norm rather than the exception across much of the country. Even the federal layer is stretched: FEMA entered the 2025 hurricane season with roughly 12% of its incident management workforce available to deploy.
The workload has not shrunk to match. A compliant emergency operations plan still requires hazard analysis, capability assessment, stakeholder coordination, annex development, and crosswalks against federal and state guidance. For a one-person shop, those hundreds of hours come directly out of the activities that determine whether a plan works when it matters: exercises, partnerships, and community engagement.
When a planner spends a hundred hours formatting annexes and cross-referencing compliance criteria, those are a hundred hours not spent building the coalition that will actually execute the plan.
Data Is Not Capability: The Human Imperative
FEMA’s Comprehensive Preparedness Guide (CPG) 101, now in Version 3.1, organizes emergency planning around a six-step process. Read the steps closely and a pattern emerges: only two of the six are primarily analytical. The rest are coordination — assembling the right people, balancing stakeholder priorities, securing approval from elected officials, and keeping the plan alive inside a community’s planning cycles. CPG 101’s whole-community doctrine makes the point explicit, directing planners to engage the public, private, and nonprofit sectors as partners rather than audiences.
Artificial intelligence excels at processing data, recognizing patterns, and synthesizing massive inputs. But a plan is only as good as the coalition built around it, and without people, it is just data. AI cannot negotiate a mutual aid agreement, read the political temperature of a county commission, or rebuild a neighborhood’s trust after a flood. Human coordination adds the context that turns a synthesized hazard picture into an executable operational strategy.
The division of labor is a design principle: machines synthesize, humans contextualize. The plan that results is both faster to produce and more grounded in the community it protects.
A Lighter Lift Across the Six Steps
This direction is no longer speculative. FEMA’s own published AI use case inventory includes the Planning Assistant for Resilient Communities (PARC), a generative AI tool intended to help SLTT governments draft hazard mitigation plan elements, and an AI-assisted resource portal to match SLTT decision-makers with recovery and resilience programs. Plan-support AI is coming to the SLTT environment whether individual agencies prepare for it or not. The question is how to integrate it deliberately. Mapped against CPG 101, the opportunities concentrate where the work is most administrative:
| CPG 101 Step | Where AI Assists | Where Humans Lead |
|---|---|---|
| 1. Form a Collaborative Planning Team | Maps stakeholder landscapes from public records; drafts engagement materials and meeting summaries. | Recruits the team, manages relationships, and resolves competing agency priorities. |
| 2. Understand the Situation | Ingests decades of hazard history, GIS layers, census shifts, and infrastructure data into a current risk snapshot. | Validates findings against local knowledge and lived community experience the data misses. |
| 3. Determine Goals & Objectives | Drafts candidate objectives traceable to the risk analysis; flags conflicts with existing plans. | Sets priorities, makes trade-offs, and aligns objectives with political and fiscal reality. |
| 4. Plan Development | Produces first-draft annexes; cross-references content against federal and state planning criteria in minutes. | Facilitates courses-of-action sessions; decides what the agency will actually commit to. |
| 5. Plan Preparation, Review & Approval | Runs compliance and consistency checks; assembles review packages and change logs. | Briefs leadership, negotiates revisions, and secures formal promulgation. |
| 6. Plan Implementation & Maintenance | Monitors data pipelines for changes that invalidate planning assumptions; drafts AAR content from exercise records. | Runs the exercises, owns corrective actions, and keeps partners engaged between updates. |
Planning for Communities That Won’t Sit Still
A static, five-year planning rewrite cycle was a reasonable convention when communities changed slowly. It is no longer sufficient for a county adding subdivisions, distribution centers, and data centers faster than its plans can be amended. AI-driven monitoring lets a small agency hold a dynamic posture that previously required a planning staff it could never afford. As new infrastructure comes online or demographics shift, continuously monitored data pipelines flag how the change alters the local risk profile, and the collaborative planning team can adjust evacuation routes, resource positioning, and public information strategies proactively — on the community’s timeline rather than the rewrite cycle’s.
The Risks Nobody Should Skip
Thought leadership on AI that ignores its failure modes is marketing. Four risk categories deserve attention before any SLTT agency puts AI inside its planning process:
1. False Confidence and Confabulation
Generative models produce fluent, plausible text that can be wrong. A fabricated shelter capacity or an invented mutual aid resource embedded in an EOP is not an inconvenience; it is an operational hazard. NIST’s generative AI guidance identifies confabulation as a defining risk of the technology, which is precisely why outputs must be verified by someone who knows the jurisdiction.
2. Data Security and Privacy
Planning data includes critical infrastructure details, vulnerable population information, and personally identifiable information. Agencies must know where a vendor’s model processes data, what it retains, and whether the arrangement satisfies applicable data handling requirements before any sensitive material enters a prompt.
3. Equity and Bias
Models trained on historical data inherit historical blind spots. Communities that were undercounted, underinsured, or underserved in past data will be underweighted in algorithmic risk pictures. The whole-community doctrine at the center of CPG 101 demands that human planners actively correct for this, not assume the model sees everyone.
4. Transparency and Public Accountability
Emergency plans are public documents, and the analysis behind them may be subject to records requests and legal scrutiny. An agency must be able to explain how a conclusion was reached. “The model said so” is not an answer that survives an after-action review or a courtroom.
None of these risks argues against adoption. They argue for governance. The NIST AI Risk Management Framework gives agencies a voluntary, widely adopted structure for doing exactly that. SLTT agencies do not need to invent a governance model; they need to borrow one and enforce it.
Human-in-the-Loop by Design
Sentinel Resilience Partners applies a strict division of responsibility designed to protect public safety operations. The principle is simple: AI drafts and analyzes; the emergency manager decides and owns. Every AI output passes through a verification layer before it touches an operational document, and nothing is published, promulgated, or briefed without a named human owner. The emergency manager remains the definitive strategic leader; the AI remains an assistant with a fast keyboard.
A Practical Adoption Roadmap
Most SLTT agencies cannot hire an AI officer, and they do not need to. A staged approach keeps risk proportional to experience:
Phase 1 — Govern First (Months 0–3). Before any tool is procured, adopt a short written AI use policy: what data may and may not enter external systems, which outputs require verification, and how AI assistance is disclosed in published documents.
Phase 2 — Pilot the Low-Risk Lifts (Months 3–9). Start where errors are cheap and visible: meeting summaries, draft after-action report sections, grant application boilerplate, plan formatting, and compliance crosswalks of existing plans.
Phase 3 — Integrate into the Planning Cycle (Months 9–18). With governance habits established, bring AI into the six-step process itself. Measure success in one currency above all others: staff hours returned to stakeholder engagement, exercises, and community presence.
Conclusion
The question facing the profession is not whether AI will change emergency planning. Federal agencies are already building plan-support AI for the SLTT audience, and the technology is already in the office through every browser. The question is whether emergency managers will govern these tools deliberately or absorb them by accident.
The planners who thrive in the next decade will not be the ones who type the fastest. They never were. The profession’s value has always lived in trust, coordination, and judgment, and those are precisely the assets AI cannot replicate and can only protect, by giving the people who hold them their time back.
Sentinel Resilience Partners is a strategic advisory firm specializing in emergency management, crisis preparedness, continuity of operations, and resilience consulting. Sentinel’s ALIGN methodology pairs CPG 101 and National Preparedness System rigor with modern analytical tooling under permanent human-in-the-loop governance. The full white paper, including references, is available as a downloadable PDF above.