Public Safety & Emergency ServicesLatest editionIssue 65
Emergency Services
A modest edition with four sources, mostly about how much evidence stands behind AI in dispatch and EMS.
Full research summary
A modest edition with four sources, mostly about how much evidence stands behind AI in dispatch and EMS. A peer-reviewed scoping review found no independent external validation in 11 EMS call-taking and dispatch model studies and no consistent benefit in two randomized trials; a Korean simulation shows an LLM matching paramedics on text-scenario triage but without field validation. A South East Coast Ambulance Service survey reports paramedics accept AI only with safeguards and retained human responsibility, and a German county announced an on-premises AI translation tool in live dispatch use, with no measured results. Fresh news since Nov. 9 was thin.
Read the edition Previous: Issue 64, November 9All Emergency Services editions
What this stream covers
Fire, EMS, 911 dispatch, emergency management, disaster readiness and response. Evaluate reliability, response time, life-safety failure modes and continuity; routine policing belongs primarily to Public Safety.
- Evidence records
- 4
- Cross-source patterns
- 2
- Also published November 10
- AnthropicAWSCampus OperationsCiscoCitizen Engagement & ServicesCrowdStrikeDellEverpureGoogleHealth & Human ServicesK–12Local GovernmentMicrosoftNetAppNVIDIAOpenAIPalo Alto NetworksPublic SafetyRed HatResearch UniversitiesServiceNowState GovernmentStudent SuccessTransportationUtility ManagementVAST Data
- Subscribe
- Atom feed for Emergency Services
What the evidence supports, and what it leaves open
2 patterns this edition will stand behind, each carried by at least two independent sources. Every one arrives with the question it has not answered.
- Reported accuracy is mostly simulated or retrospective, not field-validated
Operating questionWhat local, prospective evidence, not only model accuracy, must exist before an AI triage or dispatch aid influences live calls?
- Both clinicians and operators frame AI as support with humans keeping final authority
Operating questionHow will the agency measure whether human oversight actually catches AI errors, rather than only stating that humans decide?
Research through your lens
Every resource includes source evidence and takeaways for all three roles.
Evidence in this micro-vertical
276 resources
Follow the outcomes
276 resources across outcomes in your selection. Counts include all outcomes.
Refine by evidence type and topic
SECAmb PARA-AI: paramedics back AI in clinical care only with safeguards and want final responsibility kept with clinicians
HTN reports that SECAmb presented findings from its ten-month, mixed-methods PARA-AI project at a clinical research summit: 171 survey responses plus 14 in-depth paramedic interviews. Of respondents, 83% would use AI in clinical practice if safeguards were in place, 80% stressed clear governance, audit and accountability, 92% wanted paramedics involved in design and testing, 96% said ultimate clinical responsibility should stay with paramedics, and 83% worried about over-reliance. The trust says it turned the findings into a roadmap for the safe introduction of AI in ambulance services.
Limitations & uncertainty
Attitudes are not outcomes. The sample is 171 paramedics in one UK trust, with no reported response rate or selection method; results come second-hand via trade press from a summit presentation, not a peer-reviewed paper. Transferability to U.S. fire/EMS agencies is uncertain.
- Source
- Healio
- Published
- Healio, Nov. 8, 2026 (Scott Buzby); presented Nov. 6-9, 2026
Freiburg randomized trial: alerting cardiac-arrest volunteers by predicted arrival time beat a 600 m radius but not a 2,500 m radius
Healio reports a prospective randomized trial in Freiburg (January to July 2024) that sent 1,482 volunteers to 823 suspected cardiac arrests across three arms. A "smart alerting" algorithm that selects volunteers by estimated travel time and screens out those predicted to arrive after the ambulance was compared with fixed 600 m and 2,500 m radius alerting. Volunteers arrived before EMS 58% of the time with smart alerts, 57% with 2,500 m and 31% with 600 m. Smart alerting beat 600 m (risk difference 27 percentage points, 95% CI 19-34, P < .001) but did not differ significantly from 2,500 m. AED arrival before the ambulance was 27% (smart), 28% (2,500 m) and 11% (600 m) among volunteers reaching the scene. The system rejected 31% of responders as likely to arrive later than the ambulance, which the researchers say may reduce futile dispatches.
Limitations & uncertainty
Conference presentation reported by a trade outlet, not a peer-reviewed paper; only system performance was measured, not patient survival; randomization was by day of week rather than emergency type; German setting; no significant advantage over the 2,500 m radius.
- Source
- Discover Artificial Intelligence
- Published
- Published 7 November 2026, Volume 6, article 1817
Open-access paper: flood-cascade forecasting and resource-scheduling model is trained and tested on synthetic data, with one retrospective check on the 2021 Zhengzhou flood
The paper couples a Temporal Fusion Transformer with a spatio-temporal graph neural network to forecast how flood and heavy-rain impacts cascade between regions, then feeds prediction intervals into an integer-programming resource scheduler. On a synthetic benchmark built from EM-DAT, OpenStreetMap, WorldPop, SRTM and LISFLOOD inputs the authors report RMSE 0.071 for cascade propagation, recall 0.97, precision 0.95 and F1 0.96 under extreme node density, and a simulated response time of 5.6 s at 500 nodes. A retrospective test on the July 2021 Zhengzhou flood, with the model trained only on synthetic data, identified four of five hardest-hit regions (80%) and a cascade direction matching the documented sequence. The authors call it a proof-of-concept and state the evidence is simulation-based.
Limitations & uncertainty
Primary evidence is simulation calibrated with public data; one retrospective case with a model never trained on real data; no real-time operational validation; path-capacity degradation function unvalidated; fairness not explicitly modeled; very high reported scores arise on synthetic data.
Aspen Fire plans 2027 launch of AI-supported Seneca suppression drones after prescribed-burn tests and a limited FAA waiver
Aspen Public Radio reported on 6 November 2026 (9:02 p.m. MST) that the Aspen Fire Protection District is training to become, once staff finish training, the first U.S. fire department—and, the story says, the first in the world—to operate Seneca’s AI-supported firefighting drones independently. Reporter Halle Zander observed 24 August 2026 training in unincorporated Pitkin County. Each aircraft is described as about 10 by 12 feet, able to fly up to 20 miles empty, and designed to work in a swarm; together, five drones can hold 500 to 1,000 gallons of finished fire foam. When they arrive at a fire they use AI to calculate distance and wind speed, then release water or retardant at a specific angle and velocity. Volunteer firefighter John Hammarback called the system “pretty self-flying,” with obstacle stopping; Deputy Chief Andy Fisher wants them sent ahead to remote lightning starts. Seneca founder Stuart Landesberg compared them to a “flying Waymo,” said they will not replace air tankers on large fires, and claimed they can be “the next best thing at two orders of magnitude less cost.” Aspen Fire says it will pay about $500,000 a year over five years (about $2.5 million), funded by private donors. Charles Werner of DRONERESPONDERS noted the drones have only been tested on prescribed burns, not real wildfires, and said he does not think their size will be sufficient for what the department wants to accomplish. Aspen Fire secured an FAA waiver in late October and hopes to use the aircraft on real fires as early as next year, limited to 200 feet in Pitkin County, not over populated areas, and only with eyes on the aircraft at all times—no sending a drone blindly over a ridgeline.
Limitations & uncertainty
No measured wildfire containment results; Seneca and department claims about ease of use and cost are not independently audited here; Werner explicitly questions whether current airframes are large enough; “first in the world” is the reporter’s/department framing; related vendor stories already archived should not be treated as new performance evidence.
- Source
- Ministério da Ciência, Tecnologia e Inovação
- Published
- 2026-11-06
Brazil’s Cemaden says FloodcastingXAI operational tests improved 1-to-4-hour flood forecasts by about 40%
On 6 November 2026 (18h16), Brazil’s science ministry reported that Cemaden researchers developed FloodcastingXAI, an explainable-AI platform that ingests rain-gauge and river-level observations updated every 10 minutes to forecast water-level rise one to four hours ahead for civil-defense decision support. The ministry says the system does not replace monitoring staff or alert issuers; it adds information to technical analysis and can show which inputs most influenced a forecast. Operational tests reportedly showed gains of approximately 40% in forecast performance for one- to four-hour lead times; the release does not name the skill score or baseline behind that figure. FloodcastingXAI is a product of the iFast project (more than 60 researchers at 15 Brazilian and foreign institutions) with collaborators including Inpe, Unesp, Unifesp, and the City of São Paulo. Development used open source without external contracting. The tool is a finalist in the 30th Enap Public Sector Innovation contest (AI for the Public Good category; more than 700 entries; competing with five other projects in that category; popular vote open until 11 November; initiative no. 983), led by Leonardo Bacelar with Cemaden teammates named in the release. It previously received an AI4GOOD prize at Brazil Conference 2025. The same article mentions Cemaden’s separate GeoRisk landslide tool (up to 72-hour risk forecasts) as a related but distinct system.
Limitations & uncertainty
Ministry self-report of its research unit’s tests; the 40% figure has no named metric, geography, or independent audit in this article; hydrologic transfer to U.S. basins is unproven here; Enap finalist status is not a performance evaluation.
SAE paper: LEO satellite onboard AI aims for ~70-second intra-pass wildfire re-tasking instead of hours-long ground loops
Published Nov. 6, 2026, Qiu et al. propose Intra-Pass Autonomous Reaction (IPAR) for an agile ~800 km sun-synchronous LEO platform pairing 0.5 m optical with a forward-looking IR camera (600 km swath, 10 m nadir). Onboard AI analyzes IR in real time to detect/confirm wildfires and autonomously re-task high-resolution follow-up in the same pass, targeting response measured in seconds rather than multi-hour ground-in-the-loop latencies. Engineering budget: ~70-second window needing ~33.6° minimum IR forward look, allocating 30 s AI analysis, ≤33.6 s attitude maneuver, 10 s settle; IR GSD degrades to ~12.2 m at that angle, still argued sufficient for early detection. Closed access beyond abstract/landing content; no on-orbit flight results presented in the opened material.
Limitations & uncertainty
Engineering analysis / conference paper; full PDF paywalled in this run; Chinese platform specifics; no published field wildfire outcome metrics in the opened abstract.
- Source
- Texas A&M Stories
- Published
- Texas A&M Stories, Nov. 6, 2026; underlying paper dated May 26, 2026
Texas A&M announces CrisisSense-LLM, a disaster social-media classifier still in development and not yet deployed in an active disaster
A university press release describes an NSF-funded large language model that filters and categorizes social media posts during crises such as hurricanes into disaster-related, weather-related and humanitarian information, and further by event type, informativeness and humanitarian information type. It was tested on a benchmark of human-annotated posts from multiple disaster events. The release states the system is under development and has not been deployed during an active disaster, and the team plans an interactive AI agent for emergency professionals. The related paper (CrisisSense-LLM, DOI 10.1016/j.aei.2026.104836, May 26, 2026) was not opened, and the release gives no accuracy figures.
Limitations & uncertainty
University press release with no performance metrics; benchmark evaluation only; not deployed; paper not opened; social media data bias toward connected, English-speaking users is not discussed.
Seoul council audit: 5.078 billion-won 119 AI system scored 72/100 and documents put formal pilot call-taking on hold
At the Nov. 5, 2026 Seoul council administrative audit, councilmember Kim Jun-sung said the Fire & Disaster Headquarters' 119 AI project—about 5.078 billion won across two phases, past completion inspection and residual payment—received a comprehensive evaluation score of 72/100, and the performance measurement report listed formal pilot operation as fully postponed. Fire headquarters replied that functions other than AI directly accepting/handling 119 calls are operating and that 'full postpone' applies to some functions; the headquarters chief said he would verify facts and consider external review. Council also questioned duplicate maintenance staffing budgets and low field use of other advanced gear (quadruped robot, drone repairs).
Limitations & uncertainty
Single newspaper account of an audit dispute; full evaluation protocol and raw metrics are not published in the article; HQ contests the scope of 'postpone.' International transferability limits apply.
- Source
- National Institute of Standards and Technology
- Published
- 2026-11-05
NIST PSCR’s INSIGHT pipeline uses foundation models plus a required human reviewer to turn building scans into first-responder indoor maps
In a 5 November 2026 article, NIST’s PSCR Innovation Lab describes INSIGHT (Indoor Scene Intelligence from Geometric-Semantic Hierarchy Transfer), an automated pipeline that converts building scans into structured data for pre-incident planning, with later phases envisioned for live tracking and command. PSCR argues indoor disasters lack the satellite, GIS, GPS, and drone coverage available outdoors, while 2D blueprints often omit standpipes, shutoffs, alarm panels, and AEDs. NFPA 1660 and NFPA 950 require detailed, interoperable pre-incident data that departments now produce manually. Purpose-built 3D models “score near zero” on small, visually varied safety fixtures (exit signs, alarm panels, AEDs); PSCR says it measured that gap in a 2025 cross-dataset analysis using Point Cloud City, and that foundation models can recognize those features without task-specific training, with registered depth placing them in 3D. The pipeline is designed to follow the NIST AI Risk Management Framework by keeping a human reviewer in the loop to correct the generated graph, using independent detection paths that flag disagreement, and applying building-code bounds to implausible results—because “without proper verification, results could endanger first responders.” Outputs are intended to reach responders in a matter of seconds over existing broadband, use public data and open-source tools, follow ISO 19164, IndoorGML, and IFC naming rather than a proprietary format, and support role-filtered views for dispatch, incident command, and discipline-specific users. Current work is strictly pre-incident.
Limitations & uncertainty
This is a NIST lab news article about a pipeline under test, not a published accuracy table for INSIGHT itself; “near zero” refers to purpose-built 3D models on safety-fixture classes in a 2025 analysis, not a full INSIGHT field trial; live tracking remains future work; no timed comparison versus manual pre-plans is given beyond qualitative “seconds” delivery and “exponentially less work” for reviewers.
Dryad plans multi-month Silvaguard fire-drone pilots after XPRIZE Wildfire second place; commercialization eyed for 2028
Commercial UAV News (Nov. 4, 2026; page modified Nov. 6) interviews Dryad CEO Carsten Brinkschulte and XPRIZE's Andrea Santy on Dryad's second-place Autonomous Wildfire Response finish. Silvanet solar gas sensors (VOC/CO/PM2.5 with onboard AI) trigger Silvaguard: an IR observation drone then a suppression drone (40 L water used in Alaska; ~100 L capacity). Finals required detect-and-suppress within 10 minutes across 1,000 km² near Nenana with decoys; Dryad deployed 170 sensors (planned 1,000) after the zone narrowed. Brinkschulte says a camera they also deployed detected no fires. Silvanet is in production; Silvaguard remains prototype; company plans two–three live pilots next year (~15,000 acres, six–eight months) and commercial drone sales from 2028. XPRIZE plans a 12–15 month impact phase matching teams to end users. Prize split cited elsewhere: Anduril first, Dryad second ($800k).
Limitations & uncertainty
Company-favorable interview; Alaska results are competition conditions; camera-vs-sensor comparison is Dryad's own reading; U.S. regulatory path for heavy suppression drones not proven here.
Pano AI opens wildfire camera detections to drone partners so triangulated smoke alerts can automatically trigger Seneca, Rain, BRINC and Censys aircraft
DroneXL reports Pano AI announced a Drone Partner Program on Nov. 4 at the Red Sky Summit: with customer agreement, a confirmed AI smoke detection automatically alerts a partner drone provider with triangulated coordinates via Pano's Incident API. Launch partners are Seneca and Rain (suppression) and BRINC and Censys (camera drones); a LinkedIn graphic also showed Skydio. Seneca's drones are tied to Aspen Fire Protection District. Pano says it covers 95 million acres in 18 states and its cameras alerted on 735 vegetation fires in 2025, first known alert about 60% of the time; both are company figures. DroneXL notes availability depends on commercial partnerships and that Aspen still needed FAA and federal-land flight approvals.
Limitations & uncertainty
Vendor-stated metrics; capabilities subject to commercial partnerships; FAA and federal-land approvals unresolved for Aspen; no outcome data.
Landkreis Börde puts on-premises AI speech translation into live fire/EMS dispatch center use; the announcement reports no measured results
The Landkreis Börde press release says its integrated fire and EMS dispatch center in Haldensleben has introduced e*loquentia, an AI tool from Berlin firm e*Message that detects a caller's language, shows speech-to-text in German, and translates dispatcher replies (free text or stored text blocks) back to the caller. It says the dispatcher keeps all assessment and decisions, that all conversations are processed only on servers in the county, and that practitioner feedback is consistently positive. It gives one example, a Polish-speaking truck driver reporting a theft via 112. The tool came out of a pilot with e*Message. No accuracy, speed or error metrics are given.
Limitations & uncertainty
Self-reported announcement co-developed with the vendor; no metrics, no independent evaluation, one anecdotal example. German setting and law. Machine translation errors on locations, numbers or hazards are not discussed on the page.
Seattle Times: 2019 city privacy memo flagged equity risks of Fire Department AI on 911 medical calls, but live AI prompts launched in 2023 without public review
Public records obtained by The Seattle Times show Seattle's then chief privacy officer wrote a 2019 privacy risk analysis warning that Corti AI on the Fire Department's 911 call-taking platform could raise equity concerns tied to language, accents and disabilities, and recommended a race and social justice assessment and public disclosure. The project was framed as a quality-assurance pilot on completed calls and skipped Seattle's normal purchasing and full technology review. Since 2023 AI has listened to all 911 medical calls and prompts dispatchers on routing some callers to a nurse line. The department says human dispatchers stay in control; it has asked IT for retroactive privacy reviews, and UW Medicine is evaluating the AI and nurse line. Experts and ACLU of Washington urge suspending live prompts pending a public audit; the mayor's office promised an update on city AI use in early 2027.
Limitations & uncertainty
Single newspaper investigation; department disputes framing; no measured disparity data either way; UW Medicine evaluation scope and timing not disclosed.
JOUAV demonstrates an integrated tower-camera, autonomous VTOL patrol and suppression-drone wildfire system with AI image analysis at a Shanxi forest fire exercise
AirMed&Rescue (Nov. 3, Chloe Fox) reports that JOUAV demonstrated an uncrewed forest fire system on Sept. 28 at the Shanxi provincial forest and grassland fire skills competition and joint exercise. It combines elevated video surveillance, JOS-C700 autonomous VTOL patrols (over 120 minutes endurance, up to 50 km line-of-sight radius, visible and thermal cameras) and JOS-P1000 multirotor suppression drones (100 kg payload, four 20 kg firefighting bombs). AI-based image analysis supports assessing smoke, flames and heat sources. JOUAV states over 100 m2 effective suppression per mission, with results depending on conditions. The system is framed as complementing ground crews.
Limitations & uncertainty
Vendor-sourced specifications; demonstration at a competition, not an operational deployment; no AI performance data; China airspace and procurement context differ from the US.
Preprint: flood peak-depth model from 5,689 USGS high-water marks reaches 0.39 m error 6 hours ahead, but shows no skill on eleven held-out events
The authors assembled 5,689 USGS high-water-mark surveys from 39 flood events (2011-2025) and trained a cross-modal fusion network on terrain, geospatial foundation-model embeddings and antecedent hourly meteorological forcing. It predicts peak depth at surveyed marks with a mean absolute error of 0.39 m six hours before the peak, against 0.93 m for differencing the measured peak water level with the same terrain model. Forcing accounts for 77% of the attributed error reduction. On eleven held-out events, the model still separates flooded from rain-free conditions but no longer resolves depth at the marks (R2 = 0.03). The authors state the skill is confined to the events and regions sampled and transfer to unseen storms remains the open problem.
Limitations & uncertainty
Preprint, not peer reviewed; findings drawn from the abstract only; high-water marks are post-event surveys, so labels are sparse; no operational deployment reported.
Preprint: LLM weather guidance without alert access gave false reassurance in 37.7% of 69 real NWS alert cases; alert access removed it
The paper replays 117 frozen cases built from real National Weather Service alerts (69 alert cases, 32 no-alert controls, 16 non-U.S. locations) through five conditions of a Claude-based preparedness pipeline. One rater hand-labeled 342 alert-case responses. False reassurance, meaning calm guidance while a warning is active, was 0.0% for the production pipeline (0/66) and for the same alerts under a neutral prompt (0/69), but 37.7% (26/69) for a blind neutral prompt and 44.9% (31/69) when the blind prompt also told the model to disclose its blind spot. Disclosure instructions did not help (p = 0.40). With a neutral prompt the model raised some alarm on 87.5% of no-alert controls versus 0% under the production prompt. The author also reports NWS test products ingested as real alerts and a token budget failing on complex cases.
Limitations & uncertainty
Preprint by one author with one rater; the evaluated pipeline is the author’s own Claude-based system; a seven-category scheme was written after an initial binary pass, which the author discloses; U.S. NWS alerts only; no field deployment.
Frontiers: UAV YOLO fire/smoke detector hits lab mAP50 0.980 but authors require onboard flight tests and independent validation
Alibekkyzy et al. compare YOLOv10/11/12 on 4,200 annotated fire/smoke images; YOLOv10l achieved Precision 0.962, Recall 0.968, mAP50 0.980, mAP50-95 0.676 on workstation GPU. Simulated localization error 0.5–1.2 m (0.3–0.5 m with multi-view); modeled VLC signal loss 5–10% vs RF 15–25%. Authors state detection was workstation-based, communication/localization modeled/simulated, and further onboard UAV testing plus independent validation are required. Failure cases include lights, reflections, small distant fires and dense smoke.
Limitations & uncertainty
Internal validation only; no multi-seed statistics; VLC/localization not full hardware field tests; non-U.S. authors/setting.
Greece's civil protection ministry launches an LLM assistant on government cloud to guide natural disaster victims through state relief options
Greek Reporter (Nov. 3) reports the ministry launched a large-language-model assistant on civilprotection.gov.gr that answers general questions about disaster relief such as housing assistance, temporary housing and initial financial aid. It accepts typed or spoken Greek or English, keeps context within a conversation, lets users rate answers, and requires no identity or personal data because it does not assess individual claims. It runs on Greece's G-Cloud government infrastructure; the ministry cites GDPR, the EU AI Act and Greek Law 5321/2026, says content is checked and updated, and includes official source references when possible. The ministry cautions answers may be inaccurate and are not administrative decisions.
Limitations & uncertainty
Launch announcement only; vendor and model not named; no performance data; information-only scope.
Korean simulation: a Pre-KTAS-trained ChatGPT matched paramedics on triage accuracy (0.81 vs 0.80), but only on text scenarios with no field validation
Paramedic Papers summarizes a Korean simulation study of 100 prehospital triage scenarios (Pre-KTAS scale), each run 30 times. A Pre-KTAS-trained ChatGPT matched 10 active-duty 119 paramedics on accuracy (0.81 vs 0.80), with the lowest under-triage rate (6.1%) and highest sensitivity (0.95), while paramedics had the highest specificity (0.96) and PPV (0.93). An untrained LLM scored 0.69, significantly lower (p<.001). The summary lists caveats: simulated text scenarios, a Korean triage scale, repetition that may inflate apparent precision, and no field validation, and labels the work hypothesis-generating.
Limitations & uncertainty
Secondary summary of a paper the agent could not open; the page itself notes simulated text scenarios, a Korean triage scale, repeated runs and no field validation. Paramedic comparison group is 10 people. Peer-review status of the journal paper was not confirmed by the agent.
- Source
- Te Herenga Waka—Victoria University of Wellington
- Published
- 2026-11-03
Wellington ambulance ML recommendations improved simulated average and 90th-percentile response times; live parallel run slated for mid-2027
A 3 November 2026 Victoria University of Wellington comment by PhD researcher Jordan MacLachlan describes a machine-learning sidecar, funded by an MBIE Smart Ideas grant and supervised by Professor Yi Mei, that presents dispatch recommendations in a fraction of a second which dispatchers can accept, change, or ignore. Identical situations return identical recommendations so decisions can be audited. In late 2024, Health New Zealand asked publicly traded Norwegian firm Omda to simulate one year of de-identified historic Wellington Free Ambulance emergencies; compared with a dispatcher-style benchmark co-designed with the service, the model improved average and 90th-percentile response times across all urgency categories. That result led Health New Zealand, ACC, and Wellington Free Ambulance to support a phased real-world trial. A six-month passive parallel run on data de-identified in real time is planned to start in mid-2027, comparing every dispatcher decision with the model without affecting live dispatch. The opened page does not publish a seconds-saved figure. MacLachlan’s team has started a company, Weewoo (six people). Context cited: Hato Hone St John and Wellington Free Ambulance attended more than 640,000 incidents in the 12 months to July 2026, up more than 12% on the 12 months to July 2022; the average Wellington paramedic lasts only seven years on the road, with roughly 28% still working a decade after registration versus around 73% of doctors.
Limitations & uncertainty
First-person commentary by the researcher who founded Weewoo; Omda results are summarized without a published seconds-saved figure, confidence interval, or open methods paper on this page; simulation of historic year is not live, messy operations; New Zealand context does not automatically transfer to U.S. PSAPs.
South Florida flood night: city AI voice agent routed 227 of 298 resident calls to the EOC in a median 26 seconds
Workforce Wave's Nov. 3, 2026 case study reports that during a South Florida local emergency after Nov. 1 flooding, an existing municipal AI voice agent on the city's public line was reconfigured the same afternoon to route flood/EOC intents. From Nov. 2 ~noon to Nov. 3 10:42 PM ET (~35 hours), the agent answered 298 resident calls and connected 227 to the EOC phone bank (median 26 seconds greeting-to-handoff; ~22 seconds for yes-to-greeting callers). Ten EOC connections occurred midnight–7 AM. Other outcomes: 21 answered/redirected without transfer, 31 hung up under 20 seconds, plus small transfers to city/water/admin lines. Life-threatening callers were told to hang up and dial 911; the agent never transfers those. City, partner, and residents are unnamed. Platform reliability claims (99.69% of 36,080 agency calls YTD) are vendor-reported.
Limitations & uncertainty
Vendor-authored case study; city unnamed; no independent audit; no clinical or property-loss outcomes; 'connected' is telephony handoff, not confirmed human answer or resolved need; bilingual example is anecdotal.
Sisters-Camp Sherman Fire District launches Fire Away, a 24/7 AI chatbot that answers wildfire-prep questions from district guidance and builds cost-aware home-hardening plans
Central Oregon Daily (Nov. 2) reports that the Sisters-Camp Sherman Fire District released an AI chatbot, Fire Away, that answers questions on home hardening, defensible space, evacuation levels, livestock evacuation and other wildfire topics. Fire Marshal Jeff Puller says it walks users through steps, then produces a prioritized plan for defensible space or home hardening 'with cost in mind'. He says staff cannot always answer the phone, and the bot gives answers at midnight. Answers are based on district standard practices, and people across Central Oregon can use it. The district launched it Oct. 29 (per KTVZ) and describes it as educational, not an emergency alert service. No usage, accuracy or vendor information is given.
Limitations & uncertainty
Launch coverage only; no performance data. Single local report plus a KTVZ corroboration of launch date. Vendor and model unknown. Small rural district; transfer to larger agencies depends on content maintenance capacity.
Forest Fire Management Victoria's AI smoke-detection trial now runs 24 cameras at 14 sites, including two vendor-managed, to test reliability and timeliness alongside human lookouts
Country News (Nov. 2, Australian time; Nov. 1 in Chicago) reports that FFMVic is trialling smoke-detecting AI with cameras on fire lookout towers, including Mt Wombat near Euroa. Acting chief fire officer Aaron Kennedy says the technology complements trained fire lookout observers across FFMVic's 65 detection towers. The trial runs in four phases and assesses the reliability and timeliness of smoke-detecting AI under Victorian conditions. It now has 24 cameras at 14 sites: 22 FFMVic-managed cameras at 12 sites and two vendor-managed cameras at two more. Findings will inform future remote detection. Separately, a jointly funded plantation camera network will cover more than 1.4 million hectares.
Limitations & uncertainty
No results reported. Single local news article based on agency comments. Related to, but distinct from, the archived Oct. 26 Victorian Government plantation-camera release; the plantation network paragraph here is internally inconsistent (north-eastern pine forests vs. cameras in western and eastern Victoria). Australian context; transferability to US agencies is limited until results are published.
Korea's science ministry funds one-year rapid-response R&D: National Fire Agency digital-twin indoor 3D modeling for major fires and Korea Forest Service drone-and-AI inspection of wildfire recovery
DongA Science (Nov. 3, 12:00 KST; Nov. 2, 9 PM CST) reports that MSIT selected three second-phase projects under its new 'Pan-Governmental Rapid-Response R&D for On-Site Needs', which funds one-year projects immediately for urgent ministry needs outside regular budget cycles. They were chosen from eight demands from five agencies. The National Fire Agency will develop digital twin-based indoor 3D modeling to quickly identify a building's internal structure and firefighting facility locations, with on-site demonstrations, citing the Daejeon fire with 74 casualties and the Coupang logistics centre fire. The Korea Forest Service will develop drone-and-AI mobile inspection to verify wildfire recovery work and prevent landslides, noting about 405 wildfires this year. A third project targets illegal websites. No budgets or metrics are given.
Limitations & uncertainty
Announcement of selected projects only; no budgets, performers, metrics or results. The fire digital twin is not described as AI; AI is explicit for the forest-recovery and illegal-site projects. Korean national context; US transfer is limited by building-data availability and procurement rules. Related to but distinct from the archived Oct. 26 DongA Science EXPERT item and the Nov. 1 ETRI item.
Fraunhofer EMI RegioGuard-PRO: AI civil-protection dashboard fusing geodata for municipal crisis decisions; demonstrator in ~2 years
Fraunhofer EMI and partners are building RegioGuard-PRO, an AI-driven regional civil-protection dashboard that consolidates social, infrastructure, transport, population, environmental, building and weather data to detect hazards early, support what-if flood scenarios and offer chatbot-style retrieval over analyses. Demo centers are planned in Trier, Mainz and Wildau; a ready-to-use demonstrator is planned within two years. Project runs March 2026–March 2028 under BMFTR Practical Lighthouses of Civil Security. No field outcome metrics yet.
Limitations & uncertainty
Institute press on its own project; no measured operational outcomes; chatbot and cascade models unvalidated in this source; German context.
ETRI to show a digital federated twin for disaster prediction and an AI that analyses vital signs in ambulances to help diagnose cerebrovascular conditions, with no validation results reported
The Herald Business (Nov. 2, 10:16 KST; Nov. 1 in Chicago) reports that ETRI will show nine AIoT technologies at the 2026 AIoT International Exhibition at COEX in Seoul, opening Tuesday for three days. The flagship 'intelligent digital federated twin' links multiple digital twins to forecast conditions and assess environmental changes and the likelihood of disasters before they occur. ETRI will also show AI that analyses patients' vital signs in moving environments such as ambulances to help diagnose cerebrovascular conditions, which it expects to speed decisions by emergency medical personnel. Other exhibits include camera-and-LiDAR unmanned surveillance and livestock-barn monitoring to prevent hazards such as fire.
Limitations & uncertainty
Exhibition announcement only; no results, validation or deployment. Developer claims ('expected to support faster decision-making') are unverified. English text is AI-assisted translation. Korean research context; limited transferability. Published Nov. 2 KST (Nov. 1 in Chicago); the exhibition opens Nov. 3.
- Source
- Scandinavian Journal of Trauma, Resuscitation and Emergency Medicine
- Published
- Published 2 November 2026, Volume 34, article 170
Uppsala study: regression and gradient-boosting EMS travel-time models validated on later data explain 70-79% of ground variance but only 38-46% for helicopters
Researchers used 2022-2024 dispatch and flight-log data from Region Uppsala (24,756 priority-1 incidents, 527 with helicopter response) to adjust open-source routing estimates for ground ambulances, and great-circle distances for helicopters, with linear regression and gradient boosting. In validation data collected after the methods were finalized, the best ground models explained R2 0.79 (travel to incident) and 0.70 (travel to hospital); helicopter models reached R2 0.46 and 0.38. Mean absolute percentage error was 18-20% across techniques and intervals. Regression and gradient boosting performed similarly, with a slight edge for boosting. The authors say open-source code could help dispatchers estimate relative travel times and judge whether a helicopter saves time.
Limitations & uncertainty
Single Swedish region; priority-1 missions transported to one primary hospital only; extreme long travel times tended to be under-estimated; helicopter volume was small (527 responses); no patient-outcome or live-dispatch evaluation; Swedish road network and agency structure.
- Source
- Scandinavian Journal of Trauma, Resuscitation and Emergency Medicine
- Published
- 2026-11-02
Uppsala study validates open-source routing plus gradient-boosting and regression models to predict ground and helicopter EMS travel times for dispatch decisions
Spangler et al. (published Nov. 2, accepted early version) estimated ground EMS travel times with the open-source OSRM routing tool, then adjusted them with linear regression and gradient boosting trained on historical travel times; helicopter times used great-circle distance adjusted by historical data. On validation data collected after the methods were finalized, the best ground models explained much of the variation in drive time to scene and to hospital (R2 0.79 and 0.70), but helicopter flight times much less (R2 0.46 and 0.38). Mean average percentage error was 18-20% across all techniques and intervals. The authors release open-source code and say it could help dispatchers judge the time benefit of HEMS versus ground units.
Limitations & uncertainty
Single Swedish region; early accepted version; abstract-level inspection; helicopter predictions weak (R2 0.38-0.46); no evaluation of dispatcher decisions or patient outcomes.
Türkiye creates a General Directorate of 112 Emergency Call and Coordination and plans AI-powered decision-making systems for a service handling nearly 100 million calls a year
Daily Sabah, reporting with Anadolu Agency on Nov. 1, says a presidential decree published in the Official Gazette on Saturday (Oct. 31) created a General Directorate of 112 Emergency Call and Coordination under the Interior Ministry, replacing a department-level unit. The 112 centres in all 81 provinces become independent provincial directorates reporting to governors' offices. The directorate will oversee upgrades to the national operations centre, switchboard and communications. Authorities plan AI-powered decision-making systems and big data analysis infrastructure, plus a cybersecurity and disaster recovery centre, along with staff training, regular assessments and drills. The system handles nearly 100 million calls and more than 31 million incidents a year with about 15,000 personnel.
Limitations & uncertainty
Plans only; no AI system, vendor, budget, timeline, accuracy target or evaluation method is described. Single news report based on a government statement. The decree was published Oct. 31 and the report Nov. 1. National, centralised model; limited direct transfer to US county PSAPs.
- Source
- WIRED (republished from High Country News via Climate Desk)
- Published
- 2026-11-01
WIRED/High Country News: search-and-rescue officials describe rescues tied to chatbot trip planning, but Wyoming SAR calls fell, so any rise in rescues is unproven
WIRED (Nov. 1, by Christine Peterson; first published by High Country News) reports that three hikers rescued on Mount Shasta said their poor planning came partly from relying on Google's Gemini. A Forest Service manager also describes a visitor camping by a 'No Camping' sign because AI said she could, and a man with an AI-generated map showing features on the wrong side of a mountain. Grand Teton's superintendent says visitors reporting AI use for travel information went from zero to 26 percent. But it remains unclear whether AI use is raising rescues: Wyoming SAR calls were 312 so far in 2026, against 466 in 2025 and 377 in 2024. A Clemson researcher warns of 'cognitive surrender'.
Limitations & uncertainty
Anecdotal incidents and interview claims; no systematic data. Wyoming calls fell, which cuts against a simple 'AI causes rescues' story. The BC hiker also used Google Maps and AllTrails. This is a Nov. 1 WIRED republication of a High Country News story dated Oct. 23; neither version was in the archive. Partly outside US SLED (British Columbia).
Stream editions
Each edition carries its own synthesis and evidence ledger.
November 10, 20261 edition
November 9, 20261 edition
November 8, 20261 edition
November 7, 20261 edition
November 6, 20261 edition
November 5, 20261 edition
November 4, 20261 edition
November 3, 20261 edition