Public Safety & Emergency ServicesIssue 124 records
Emergency Services
Four newly catalogued sources after the 2026-09-16 edition: Sept 16 Muon/EFA first-light imagery from the operational FireSat constellation (still commissioning; Early Adopter data ≥2× daily by end-2026) without independent false-alarm outcomes; Fox 8’s Jul 16 report that Orleans Parish uses AI on 911 for geofenced duplicate crash surges with operator-claimed early listening QA and explicitly anecdotal life-saving evidence; Signal Akron’s Sept 9 Documenters record of Ava non-emergency AI with ~39k fully automated calls and resident spillover questions; and CISA SAFECOM/NCSWIC’s AI-in-ECCs infographic requiring evaluation of operability, interoperability, security, and resilience. Two patterns separate automation volume from life-safety proof and bind federal evaluation/cyber gates to local public-trust scrutiny. No new controlled U.S. response-time or clinical-benefit trial dated after 2026-09-16 is claimed.
- Evidence records
- 4
- Cross-source patterns
- 2
- Evidence classes
- 2 independent reporting1 vendor claim1 standards or public-body guidance
- Outcomes
- 2 mixed1 emerging1 cautionary
- Source freshness
- 2 new this fortnight1 recent1 older, newly relevant
- Research completed
- 2026-09-18
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Treat PSAP AI call-share and automation counts as capacity signals, not life-safety proof
Fox 8’s OPCD story publishes high duplicate-crash AI share and early zero FP/FN listening claims while stating life-saving evidence would be anecdotal, and Akron’s Documenters record cites tens of thousands of automated non-emergency calls with a 7% callback figure whose period is unclear. Together they show agencies can quantify automation volume without proving clinical or mortality benefit. SLED buyers should demand ongoing false-negative and spillover audits before treating counts as staffing or safety outcomes.
Operating questionWhich ongoing false-negative, callback, and 911-spillover metrics with dated denominators will be published before AI call-share is treated as a life-safety or FTE result?
Supporting evidenceOrleans Parish Communications District (OPCD) / Carbyne (Axon 911)Akron Combined Communications Center
Require operability, cyber, and public-trust evaluation before scaling ECC AI
CISA SAFECOM/NCSWIC tells ECCs that AI adoption needs significant evaluation of operability, interoperability, security, and resilience plus COOP-linked controls, while Akron residents pressed dispatchers on AI dislike, wait times, and whether non-emergency AI pushes callers to 911. Federal checklist expectations and local trust friction align: do not scale from trial anecdotes without implementation plans, spillover monitoring, and public scope. This is a control pattern, not evidence that either jurisdiction’s tool fails clinically.
Operating questionBefore expanding ECC AI, can the agency show a written AI implementation plan, cyber/COOP risk assessment, and public metrics addressing spillover and trust?
Supporting evidenceCISA SAFECOM / NCSWIC (NG911 Working Group)Akron Combined Communications Center
Evidence ledger
Full record · every source keeps its link and limitations
Muon Space and Earth Fire Alliance release first light and active-fire imagery from operational FireSat constellation
On 2026-09-16 Muon Space and Earth Fire Alliance released first-light and active-fire imagery from the three operational FireSat satellites launched July 7, stating MWIR can detect fires as small as 5×5 m, showing multi-day Big Grass Fire progression, and saying satellites remain in commissioning with Early Adopter data at least twice daily by end of 2026—without publishing independent false-alarm or responder outcome metrics.
Why it matters, evidence and limitations
- Why it matters
- Why now: published 2026-09-16 after the last completed emergency-services edition catalogued Google Research’s FireSat AI interview (different URL); advances from method claims to operational constellation imagery and an Early Adopter delivery timeline relevant to U.S. fire agencies.
- Evidence and measured results
- Press release dates Sept 16, 2026; July 7 launch; ~3 weeks post-launch imagery across three continents; 5×5 m MWIR sensitivity claim; Big Grass Fire ~290,000 acres example; commissioning status; Early Adopter ≥2× daily by end-2026; hourly revisit target 2029 / 20-minute early 2030s; funders include Google.org and Bezos Earth Fund.
- Limitations and uncertainty
- Vendor/partner release; no independent audit of detection precision, false-alarm rate, or firefighter decision outcomes; satellites still commissioning; AI contribution not quantified in this page.
Orleans Parish 911 uses AI to answer surge motor-vehicle-accident calls near known crashes
Fox 8 (published Jul. 16, 2026) reports OPCD uses Carbyne-developed AI to answer 911 callers within ~200 meters of a known motor-vehicle accident when humans are busy, with Fasold claiming ~15–18 of every 20 crash calls handled by AI, first-three-months 100% listening QA with no false positives/negatives, ~1 hour/day staff productivity, ~$600k/year contract including AI—while stating life-saving evidence would be anecdotal and a Tulane professor warns a single miss is life-critical.
Why it matters, evidence and limitations
- Why it matters
- Why now: not previously archived; newly relevant after 2026-09-15/16 editions focused on non-emergency-only AI—documents U.S. emergency-path (911) AI for redundant crash surges with human transfer for other emergencies and operator-claimed listening QA.
- Evidence and measured results
- Article quotes Fasold on 200m geofence, queue relief, 15–18/20 crash-call share, 100% listen for first three months with no false positives/negatives, +1 hour/day productivity, staffing gaps (42 vs 64 call takers), ~$600k contract now branded Axon 911; Lalka life-safety caution; Fasold says showy life-save examples unavailable.
- Limitations and uncertainty
- Local journalism relying on OPCD estimates; no public independent audit table; life-saving claims explicitly anecdotal; policing and fire/EMS share the same PSAP—stream relevance is dispatch capacity and emergency triage controls.
Akron dispatchers tell residents Ava AI handles non-emergency calls; ~39k fully automated
Signal Akron Documenters (published Sept 9, 2026) record an Aug 26 Ward 7 meeting where Akron dispatch said Ava AI is used for non-emergency—not 911—routing since May 2025 (still trial), citing 84,149 AI-handled calls with nearly 39,000 fully automated and 1,200/17,322 (7%) needing callbacks, while residents questioned wait times and whether AI pushed people to dial 911.
Why it matters, evidence and limitations
- Why it matters
- Why now: Sept 9 local measured non-emergency AI disclosure after the 2026-09-16 La Vergne AVA deployment edition—adds public-trust scrutiny plus callback-rate figures absent from many launch stories.
- Evidence and measured results
- Documenters notes: 62 workers; 330,000 calls/year; Ava since May 2025 trial; not on 911; 84,149 AI-handled; ~39,000 fully automated; 17,322 completed by AI agents with 1,200 (7%) callbacks (period unclear); frustration/cursing escalation to agent; residents asked about 911 spillover; presenters acknowledge dislike of AI.
- Limitations and uncertainty
- Documenters paraphrased meeting notes (not investigative audit); callback metric period unclear; combined center includes policing workloads—relevance for stream is fire/EMS-serving dispatch capacity and non-emergency AI controls.
SAFECOM/NCSWIC Artificial Intelligence in Emergency Communications Centers infographic
CISA SAFECOM/NCSWIC’s AI in ECCs infographic (PDF dated as of January 2026; news page revised April 23, 2026) frames AI as a decision-support tool for triage, translation, transcription, QA, and sensors, while stating incorporation requires significant evaluation of operability, interoperability, security, and resilience, and listing COOP risk assessment, AI implementation plan, data inventory, need-to-know access, and encryption controls—including deepfake/spoofing threats.
Why it matters, evidence and limitations
- Why it matters
- Why now: not previously archived; newly relevant after September 2026 non-emergency and emergency-path PSAP AI deployments (La Vergne, Akron, New Orleans) that need a federal evaluation/cyber checklist rather than launch metrics alone.
- Evidence and measured results
- PDF states decision-support framing; evaluation of operability/interop/security/resilience; examples (Charleston admin AI; Jefferson County up to 40% admin calls; Orleans Parish translation 70% faster claim; Nashville transcription; Hamilton County QA keywords; wildfire sensors); cybersecurity section with COOP risk assessment, AI implementation plan, data inventory, encryption; “As of January 2026.” News page: released May 20, 2025; revised April 23, 2026.
- Limitations and uncertainty
- Infographic guidance with illustrative agency anecdotes; not a controlled evaluation of outcomes; some example metrics (e.g., 70% faster translation) are unsourced in the PDF itself.
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.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.