{"resourceId":"nola-gambit-911-ai-scope-20260812","versions":[{"version":"external-b402a23a3ae61014d9304352f01e59fd38e8ff9e0eeba6f7c69674ce0a5a5e3d","resource":{"id":"nola-gambit-911-ai-scope-20260812","title":"Gambit: New Orleans 911 AI limited to near-incident crash callbacks; cannot dispatch or prioritize","organization":"Orleans Parish Communications District (OPCD)","sector":"911 / PSAP call handling","geography":"New Orleans / Orleans Parish, Louisiana, United States","publishedAt":"2026-08-12","publicationDate":"2026-08-12","eventDate":"2026-08-12","sourceName":"Gambit (nola.com)","sourceLabel":"Independent local reporting","sourceUrl":"https://www.nola.com/gambit/news/the_latest/what-we-know-about-how-new-orleans-is-and-isnt-using-ai-for-911-calls/article_d9d87cc7-7c3b-4fc0-a004-d01ad4891bf5.html","evidenceClass":"independent-reporting","outcomeClass":"emerging","topics":["operating-model","accessibility-workforce","governance-procurement"],"finding":"Gambit's Aug 12 2026 interview with OPCD Executive Director Karl Fasold clarifies that New Orleans' 911 AI (in use since 2023 for a narrow use case) answers only callers within about 200 meters of a known car accident, is not a large language model like ChatGPT, and cannot end calls, dispatch responders, or prioritize calls—while a limited 311 AI began more recently. Why now: the July fox8live item is already archived; this local primary interview supplies the operational boundary language SLED PSAPs need when vendors market 'AI answers 911,' and pairs with post–2026-09-19 scrutiny of unverified accuracy claims.","sledRelevance":"U.S. local PSAP deployment of narrowly scoped duplicate-incident AI with explicit non-dispatch limits—directly relevant to SLED 911 staffing-gap procurements.","evidence":"Opened Gambit HTML dated Aug 12, 2026 (editor's note correcting 911 AI start to 2023 vs 311 AI). Fasold quotes and statements that the system cannot end calls, dispatch, or prioritize; ~200-meter proximity trigger; not an LLM; dialect/street-name training cited in surrounding coverage context.","architectureImplications":"Geofenced duplicate-call classifier in front of human queue; hard escalate-to-human on uncertainty or non-match; keep CAD priority and unit recommendation entirely human.","governanceImplications":"Publish caller-facing disclosure, use-case boundary SOPs, and public FAQs distinguishing 911 AI from 311 AI; require board/oversight briefings before any expansion beyond crash duplicates.","securityPrivacyImplications":"Non-LLM local training claim reduces outbound LLM exfiltration risk but still requires recording, retention, and audit controls equivalent to human-handled 911 audio.","caveats":"Single local interview piece; performance metrics in this article are qualitative; does not independently audit error rates.","streamIds":["emergency-services"],"roles":{"sales":"Frame New Orleans as a staffing-relief pattern for duplicate crash surges—not a general AI call-taker. Buyers should hear Fasold's limits: no ending calls, no dispatch, no priority setting, and not an LLM. Pitch a bounded geofenced duplicate-incident pilot with public disclosure and a written expansion freeze until audited error rates exist, refusing any claim that AI is 'answering 911' for medical or violent emergencies.","engineering":"Fit is a proximity-and-intent classifier that intercepts likely duplicate crash reports when humans are busy, then escalates on uncertainty. Integrate with ANI/ALI location, open-incident buffers (~200 m), and immediate human transfer paths. Prerequisites: accurate incident geolocation, call-recording audit, and local speech tuning. Constraints: accents, covert callers, and multi-incident scenes. Security: keep audio in PSAP-controlled storage; avoid general-purpose LLM endpoints. Proof of value: duplicate-call deflection rate with sampled human review of every automated close.","delivery":"Work covers use-case SOP, queue integration, telecommunicator training on takeover cues, public communications, and a monitored launch window with 100% review before sampling. Dependencies: OPCD-like authority approval, telephony/CAD vendors, and legal/privacy review. Ownership remains the PSAP director. Skills adoption stresses treating AI closes as administrative, not clinical. Governance checkpoints: disclosure, expansion change-control, and audit sampling plan. Acceptance: zero automated closes outside the crash-duplicate rule; documented human path always available. Risks: public trust loss if scope creeps or disclosure lags."},"roleDetails":{"sales":{"customerProblem":"PSAPs need surge relief on duplicate crash calls without letting AI handle true emergencies.","stakeholders":["PSAP directors","Mayors/parish boards","Telecommunicators","EMS/fire chiefs","Privacy counsel"],"discoveryQuestions":["What exact incident types may the AI close?","Is caller disclosure published?","What is the human-transfer latency?","Who approves scope expansion?"],"valueHypothesis":"If geofenced duplicate handling cuts redundant queue time without miscloses, centers may fund a narrowly scoped module.","potentialEngagement":"Scope workshop, disclosure package, and 90-day monitored duplicate-crash pilot.","unsupportedClaims":["AI answering all 911 calls","Autonomous dispatch","Guaranteed safety for medical calls","LLM-grade conversational triage"]},"engineering":{"fit":"Geofenced duplicate-crash call deflection with mandatory human escalation paths.","architectureIntegration":"Telephony front-end tied to open CAD incidents; no write access to unit recommendation or priority fields.","prerequisites":["Reliable caller location","Open-incident geobuffer","Recording/audit pipeline","Local speech adaptation"],"constraints":["Multi-crash scenes","Involved-party callers","Accent/noise","Staffing to review audits"],"security":"Retain audio under existing 911 controls; prohibit training export to consumer LLM APIs.","proofOfValue":["Duplicate deflection count with human audit","Transfer success rate","Zero priority/dispatch actions by AI"]},"delivery":{"work":"SOP, integration, training, public FAQ, monitored launch, audit sampling cadence.","dependencies":["Board approval","Vendor telephony hooks","PIO disclosure","QA unit capacity"],"ownership":"PSAP executive director.","skillsAdoption":"Train staff to reclaim calls instantly and to explain AI scope to the public.","governanceCheckpoints":["Pre-launch disclosure","Weekly audit in month one","Formal vote before new use cases"],"acceptanceCriteria":["Cannot dispatch/prioritize/end without human","Published scope FAQ","Audit log of automated closes"],"risks":["Scope creep to medical calls","Disclosure backlash","Misclose of involved-party callers"]}},"retrievedAt":"2026-09-20T22:09:51-05:00","enrichedAt":"2026-09-20T22:09:51-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Positions AI as administrative surge relief so telecommunicators stay on true emergencies; still requires workforce capacity for audits and rapid human takeover.","procurementImplications":"Contract for enumerated use cases only, audit rights, and change-control on expansion; avoid LLM-as-a-service clauses that export 911 audio.","operatingModelImplications":"Splits queue into AI-eligible duplicates vs human-only emergencies; 311 AI is a separate track and must not be conflated in SOPs.","sourceVerification":{"responseId":"curl-nola-gambit-20260812","openedUrl":"https://www.nola.com/gambit/news/the_latest/what-we-know-about-how-new-orleans-is-and-isnt-using-ai-for-911-calls/article_d9d87cc7-7c3b-4fc0-a004-d01ad4891bf5.html","referenceExcerpt":"it is limited to relatively basic functions and cannot end calls, dispatch responders or prioritize calls","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}