Risk, Crisis and Emergency Communication Weekly

How to tell whether a warning or crisis message produced action, not just reach.

Coverage period: 08:00 on 10 August 2026 to 07:59 on 17 August 2026, Europe/Lisbon time. This edition includes only material first published or formally released in that window.

This week in one paragraph

This week's material shifts attention from message approval to handoff measurement. In the Democratic Republic of the Congo, WHO and its partners are treating community-generated evidence as part of outbreak intelligence. In the United Kingdom, the most extensive use of the Emergency Alert system for an active incident creates a live test of reach, proportionality, opt-outs and emergency-service burden. A small Trinidad and Tobago study shows what senior public-health leaders want from AI-enabled warning systems, while a Hawaii storm page shows that a warning is also a service directory. The practical question is what the recipient could do next, and what evidence shows that the route worked.

Three developments to prioritise

1. Community evidence becomes part of the outbreak picture

Published: 11 August 2026 for WHO's community-evidence funding announcement; 14 August 2026 for WHO Disease Outbreak News 615 and the IFRC community-trust account.
Region and sector: Democratic Republic of the Congo and the wider African Region; outbreak response, public health, community engagement and humanitarian operations.
Evidence status: WHO official outbreak bulletin and departmental update, plus an IFRC official press release. The sources establish the situation and the response design. They do not yet show that a particular message caused a measured behaviour change.
What changed: WHO's 14 August bulletin reported 4,665 confirmed cases and 2,184 deaths in the DRC as of 12 August, with spread across 54 health zones in six provinces. WHO also reported that the week from 3 to 9 August had the highest weekly numbers so far: 579 cases and 304 deaths. 1
WHO said part of the increase reflects stronger surveillance and testing, but that most reflects expansion of the outbreak. It highlighted local authorities, trusted community leaders and closer work with communities. It also recorded 12 attacks on health care since the emergency was declared in May. 1
The International Federation of Red Cross and Red Crescent Societies said on 14 August that volunteers were listening to community concerns, encouraging early care and helping people accept surveillance, contact tracing, infection prevention and safe burials. Its claim is operational and experience-based. It is not an independent evaluation of which messages changed behaviour. 2
On 11 August, WHO announced a US$3 million Wellcome grant for rapid assessments of community perceptions, care-seeking, response to public-health measures, social and economic effects, barriers to care and trust. 3
Professional implication: Treat rumour, trust and access evidence as operational intelligence. Record which finding changed a service, message, messenger, burial arrangement, referral route or security decision.
Limit: The funding announcement describes an approach, not an impact evaluation. IFRC's account of trusted volunteers is experience-based and does not isolate message effects. The WHO case count is a 12 August surveillance snapshot rather than a live count for publication morning.

2. The UK's wildfire alert spends credibility as well as reach

Published: 14 August 2026 for the government announcement; 15 August 2026 for an independent report on public response.
Region and sector: England and Wales; public warning, wildfire preparedness and emergency services.
Evidence status: The alert and response measures are confirmed by a GOV.UK press release. Public reaction and reported 999 call pressures come from a secondary Telegraph report and remain unverified as national totals.
What changed: The UK government sent an Emergency Alert across England and Wales from approximately 19:00, telling people not to light barbecues, campfires or fireworks and to report fires by calling 999. The release called it the most extensive use of the system for an active incident. It also announced emergency funding, military support in South Wales and a temporary nationwide ban on selling disposable barbecues. 4
The Joint Research Centre reported on 12 August that drought affected half of the European Union and the United Kingdom in July, with 9% at alert level. By 5 August, recorded burnt area in the European Union had passed 505,000 hectares. This is official analytical context, not a measurement of the alert's effect. 5
A 15 August Telegraph report described requests to disable the alerts and exceptionally high 999 call volumes reported by some fire services. No official national delivery, opt-out or call-volume dataset was available in the material reviewed. 6
Professional implication: Evaluate broad alerts on five measures: delivery, comprehension, protective action, burden on emergency services and opt-out behaviour. A strong signal can still weaken future warning capacity if recipients experience it as disproportionate.
Limit: The government source does not publish delivery or behaviour statistics. The independent report is a cautionary signal, not proof of a national backlash or a causal change in emergency calls.

3. AI early warning must prove the handoff to a human service

Published: 13 August 2026.
Region and sector: Trinidad and Tobago; public health, preparedness and AI-enabled early warning.
Evidence status: Brief research report in Frontiers in Public Health. It is a small, descriptive, single-country study based on self-reported views.
What changed: Six of nine county medical officers identified infectious disease and flooding as priority applications for AI-enabled early warning systems. Five valued dashboards and five valued integration with emergency services. Three identified automated alerts and three identified heat-related risks. All six prioritised equity, especially for underserved populations, and human oversight remained important. 7
The main barriers were budget constraints, limited technical capacity and data challenges. The authors caution that the six respondents, single-country setting and brief open-ended responses do not support broad generalisation. 7
Professional implication: Test the human route around the model. Give a health director, emergency operator and community intermediary the same output. Ask what each thinks it means, what action they are authorised to take, which group may be missing and who can stop or correct the alert.
Limit: The paper measures perceived priorities, not alert accuracy, response time or health outcomes.

Cases and operational detail

Hawaii: the warning page as a service directory

The Hawaii Emergency Management Agency's Tropical Storm Lala update was last updated on 16 August. It kept a tropical-storm warning for Kauai County, flood watches for three counties and statewide brown-water advisories. It linked shelter information, county alert registration, road closures, utility outage maps, forecasts and radio stations, including Japanese-language and multicultural radio options on Oahu. 8

The lesson is not that a long page is automatically good. The page earns its length if a resident can find an open shelter, a fallback channel and a safe next action without searching across contradictory pages. Test it as a service directory under changing conditions.

Colombia: uncertainty is part of the public message

An IFRC update published on 11 August described continuing search and rescue after a magnitude 7.4 earthquake in western Colombia. It said casualty and damage figures remained preliminary because access to Choco was difficult and roads and airports were disrupted. 9

A changing incident needs a clear uncertainty boundary. Communicators should state what is known, what is being assessed, when the next update is expected and what people should do while the figures change.

What to test next

Draw one warning or outbreak message from the sensing team to the person expected to act. Add the approving authority, translator, carrier, employer or school, community intermediary and service provider.
At each handoff, record the maximum acceptable delay, fallback channel, correction authority and evidence that the next actor understood the instruction.
Run three stress cases: the recipient has no transport; the named service has no capacity; or the AI output misses a mobile or marginalised population.
After the exercise, record what changed because of the evidence: a service, message, language, authority, channel or automated output. A listening process that changes nothing is documentation, not intelligence.

Watchlist

New WHO Bundibugyo data, community-evidence findings and cross-border preparedness.
UK delivery statistics, opt-out data and evidence of how the wildfire alert affected behaviour.
An operational pilot following the Trinidad and Tobago AI early-warning study.
Follow-up from Colombia as damage assessments stabilise, and maintenance evidence from Caribbean and Pacific warning work.

Editorial transparency

I excluded undated local warning pages, promotional material and studies whose in-window date could not be proved. UK alert facts are attributed to the government; the reported public-response concerns are attributed to the Telegraph. Ebola figures are WHO's 12 August surveillance snapshot. No equally strong, date-verifiable primary item from the Middle East made the final selection. The selected material spans Africa, Europe, the Caribbean, the Pacific and Latin America, but publication output was not evenly distributed across regions.

The episode was researched and scripted with AI assistance. Editorial review checked the publication window, opened the cited pages, searched the published history for overlap and kept evidence status and limitations explicit. The narration is synthetic.