Public Safety & Emergency ServicesIssue 093 records
Emergency Services
Catch-up edition after the missed 2026-09-14 research slot. Three newly catalogued sources: a Sept 14 NSF-funded UB REKINDLE project planning geospatial AI for cascading WUI wildfire hazards; Sept 14 KUTV reporting that Weber 911's AI handles about 10% of non-emergency calls amid trust/transparency concerns; and Seattle's Aug 31 Corti EMS documentation Privacy/AI Impact Assessment with multi-step human validation and patient opt-out—newly relevant after recent Seattle 911 AI scrutiny. Two patterns separate disclosed human accountability from automation ambition, and treat pilots/grants as non-outcome evidence. No new controlled U.S. response-time or clinical-benefit trial dated after 2026-09-13 is claimed.
- Evidence records
- 3
- Cross-source patterns
- 2
- Evidence classes
- 1 academic research1 independent reporting1 standards or public-body guidance
- Outcomes
- 2 emerging1 mixed
- Source freshness
- 2 new this fortnight1 recent
- Research completed
- 2026-09-15
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Disclose emergency AI use and keep humans accountable for the official record
Weber's AI non-emergency answering raises public trust questions when consent and transparency are weak, while Seattle's documentation PIA requires patient notice, opt-out, and multi-step EMT attestation before chart entry. Different tasks share a governance need: say when AI is present and keep an identifiable human responsible for the operational record. This is a control pattern, not proof that notice alone improves clinical outcomes.
Operating questionBefore expanding 911 or EMS AI, can the agency show public notice and a named human owner for every AI-influenced record or routing decision?
Supporting evidenceWeber Area 911 / KUTV reportingCity of Seattle / Seattle Fire Department
Treat call-share goals and research awards as capacity signals, not life-safety proof
Weber reports about 10% AI call share and an ambition toward 100% without published misroute metrics; UB announces NSF-funded cascading-wildfire AI without measured detection or response gains. Volume, funding, and method plans are useful signals of intent but do not establish safer or faster emergency outcomes. Require local outcome gates before equating automation share or grant dollars with mission success.
Operating questionWhich audited safety or response metric must improve before AI volume or research prototypes are counted as operational success?
Supporting evidenceWeber Area 911 / KUTV reportingUniversity at Buffalo (with UC Berkeley, Northeastern University, DOE PNNL)
Evidence ledger
Full record · every source keeps its link and limitations
NSF REKINDLE project plans geospatial AI for cascading wildfire and lifeline risks
On 2026-09-14 UB announced a $2 million NSF FIRE-WUI award (REKINDLE, Sept 2026 through Aug 2029) to combine physics-informed geospatial AI hazard mapping, machine-learning models of cascading power/water/transport failures, and social-science methods, aiming to inform emergency managers' investment and recovery priorities in WUI communities—without reporting measured operational results yet.
Why it matters, evidence and limitations
- Why it matters
- Why now: published the edition date and after the 2026-09-13 last run; directly targets emergency management decision support for wildfire cascading hazards, a core Emergency Services stream need, while remaining clearly pre-deployment research.
- Evidence and measured results
- Release states award size, NSF FIRE program framing, partner institutions, AI/ML method intents, LA/Montecito case studies, stakeholder engagement with agencies/utilities/communities, and education/workforce components. No accuracy, lead-time, or lives/property metrics are provided.
- Limitations and uncertainty
- Grant announcement only; no validated model performance; California case studies may not transfer; not evidence of improved dispatch, suppression, or casualty outcomes.
Weber 911 AI handles about 10% of non-emergency calls as trust concerns rise
KUTV's 2026-09-14 report states Weber 911 is using an AI program on non-emergency calls so humans can focus on emergencies; the director says capability improved over roughly 18 months, AI currently handles about 10% of their calls, and a near-term goal of 100% is discussed, while a UVU policy expert warns that lacking transparency and consent can stall even well-intentioned deployments.
Why it matters, evidence and limitations
- Why it matters
- Why now: published 2026-09-14 after the last completed Emergency Services run; provides a rare attributed U.S. PSAP AI volume claim plus public-trust constraints for non-emergency 911 automation—inside this stream's dispatch boundary (Flock/ALPR content in the same article is not the selected finding).
- Evidence and measured results
- Direct quotes from Weber 911 leadership on capability change, 10% current share, and 100% ambition; Amacher quotes on nuance, public good, and consent/transparency as the core trust problem; public-reaction soundbite calling AI non-emergency answering dystopian. No independent accuracy audit is presented.
- Limitations and uncertainty
- Single local news story; 10%/100% figures are operator claims; no published misroute or abandonment metrics; article also covers policing cameras outside this finding's scope.
Seattle Fire Corti EMS documentation pilot emphasizes HITL review and patient notice
Seattle's Aug 31, 2026 Privacy/AI Impact Assessment describes a multi-month Corti Assistant pilot in which transcription and an in-app LLM surface medically relevant fields for EMT/paramedic review before manual paste into ESO, with patient notice, opt-out, no ESO integration, stated no vendor training on Seattle data, and multiple human validation steps—documentation support, not autonomous triage.
Why it matters, evidence and limitations
- Why it matters
- Why now: newly relevant after September 2026 public scrutiny (already archived) of Seattle Fire AI on 911 medical calls; this PIA is a distinct, previously unarchived governance artifact for a field-documentation pilot that makes notice and HITL controls explicit, informing SLED delivery patterns for EMS AI.
- Evidence and measured results
- PIA sections detail scope (handwritten/typed notes vs transcription+LLM surfacing), validation sequence, consent/notice language, HIPAA/RCW framing, cloud processing claims, retention alignment, and explicit statement that the system produces recommendations and cannot act without human intervention. Anticipated benefits are not measured in the document.
- Limitations and uncertainty
- Self-assessment not independent audit; benefits unverified; does not evaluate Corti live triage prompting; pilot parameters may change.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.