Transcript: Episode 17

And I'm Sarah.

This one is unusually actionable. The fund explicitly names two focus areas: Resilience, which covers AI forecasting to help communities prepare for and recover from crises, and Health, meaning public health emergency systems. Grants run $1 million to $3 million, and they come with Google engineering mentorship through a technical accelerator. The application deadline is April 3rd. If you are an NGO, academic lab, or social enterprise building AI tools for disaster preparedness or climate adaptation — apply now. The Resilience focus area was written for this audience.

Indonesia is one of the most disaster-exposed countries on earth — floods, earthquakes, volcanic eruptions, cyclones. What's significant here is this is not a pilot. This is live, national-scale deployment at the meteorological agency that generates official early warnings for hundreds of millions of people. The system combines Weathernews' high-resolution proprietary data with Google's machine learning models to produce earlier, more precise alerts for riverine flooding and cyclone tracks. For anyone building AI-enhanced preparedness systems in Southeast Asia, this is the model to study — it shows how commercial AI can be operationalized through national disaster management agencies.

Compliance timelines are live. The law uses a risk-based classification system similar to the EU AI Act. High-risk AI systems — and that may include AI-powered needs assessments, biometric data tools, and medical diagnostics — require conformity assessments before deployment in Vietnam. There's also mandatory labeling of AI-generated images and video, and a requirement that any AI system serving humans be "human-centric." If your organization runs AI programs in Vietnam, audit them now. This isn't theoretical — enforcement is active as of today.

Cash programming is one of the most effective tools in humanitarian response — but verification and fraud risk are persistent challenges. This platform uses blockchain rails with item-level controls, meaning aid organizations can restrict digital vouchers to specific categories of goods — food, medical supplies — and get an immutable audit trail of how every dollar was spent. Both companies are presenting at Disasters Expo USA in Miami on March 4th and 5th. The timing is relevant: as federal disaster aid infrastructure has become less predictable, this kind of private-sector cash programming infrastructure is designed to fill the gap at scale.

Today's AI tools are largely reactive and session-based. You ask a question, you get an answer, and the conversation resets. The shift to stateful, persistent agents is a different capability class. Imagine an AI agent that tracks a relief supply chain across a four-week disaster response — maintaining operational memory, coordinating with multiple vendors and field teams, flagging anomalies without needing to be re-prompted constantly. That is what agentic infrastructure makes possible. We are not there yet in humanitarian deployment, but the underlying infrastructure is being built now. Practitioners should start thinking about where persistent agency would have the most impact in their workflows.

And 36% have no plans to create one. That number should worry anyone in humanitarian communications. A deepfake of a government official announcing a false evacuation order, or fabricated footage claiming disaster conditions in a specific area, could cause real harm during an active crisis. This is not a future risk — it is a current one. The call to action is simple: build a deepfake response playbook before the next major crisis, not during it. The survey also found critical gaps in AI literacy, data analytics, and disinformation response capabilities across crisis comms teams globally. There is significant ground to cover.

Stay ready.