{"resourceId":"rapidsos-apco-state-of-ai-public-safety-survey-202610","versions":[{"version":"external-4c8e0ae5effbeb14653850c7aaf6ddd44adef7c1d2d6c0e1dfa839444825f466","resource":{"id":"rapidsos-apco-state-of-ai-public-safety-survey-202610","title":"RapidSOS×APCO survey: 42% of ECCs deploy or pilot AI, but only ~8% completed any AI risk assessment and high-stakes trust stays low","organization":"RapidSOS / APCO International","sector":"911 / Emergency Communications Centers","geography":"United States (740 respondents; 53 states/regions represented)","publishedAt":"October 2026","publicationDate":"2026-10-01","eventDate":null,"sourceName":"RapidSOS × APCO International State of AI in Public Safety Survey 2026","sourceLabel":"Joint industry–association survey of 740 ECC directors, supervisors, and telecommunicators","sourceUrl":"https://info.rapidsos.com/hubfs/RapidSOSxAPCO-International-State-of-AI-in-Public-Safety-Survey-2026.pdf","evidenceClass":"public-sector-association","outcomeClass":"cautionary","topics":["governance-procurement","operating-model","data-security","accessibility-workforce"],"finding":"October 2026 RapidSOS×APCO report (fielded July 31–Sept 17, 2026; n=740) finds 27.0% of ECCs actively deploy AI and 14.5% are piloting (42% combined), with real-time transcription the top capability (~38.9%). Only 3.2% report an established formal AI risk-assessment framework and 4.9% an informal one (~8% any assessment); 49.7% say no. Mean trust that AI can reliably support (not replace) human judgment in high-stakes decisions is 2.5/5; only 18% rate trust at 4–5. Accuracy/reliability is the top adoption barrier (75.3% top-3). Liability for AI-assisted decisions is the top legal concern (75.4%). 40.1% have no AI policy and no plans. Why now: pairs today's vendor EOC routing metrics with national ECC self-report that governance and trust lag adoption.","sledRelevance":"Core U.S. 911/fire-EMS dispatch governance signal: adoption is spreading while documented risk assessment, policy, and high-stakes trust remain scarce—directly frames sales and delivery of emergency-services AI.","evidence":"Opened PDF report; methodology states 740 respondents, no incentives; figures cited from Q4, Q16, Q29, Q9, Q34 as printed.","architectureImplications":"Prioritize human-in-the-loop assistive uses (transcription, translation, QA) with explicit non-automation of high-stakes judgment until risk assessments and accuracy evaluations exist.","governanceImplications":"Agencies adopting AI should complete documented risk assessments, assign legal oversight, and publish community notices—areas where the survey shows large gaps.","securityPrivacyImplications":"Respondents rate cyber/data security and privacy among material risks; liability and CJIS/privacy evaluation difficulty are highlighted as barriers.","caveats":"Vendor-co-branded survey (RapidSOS) with APCO; self-selected respondents; not a controlled performance study of any product; some policing-capable ECCs included though questions are ECC/911 focused; October publication is slightly before the Nov. 5 last run but newly paired here with fresh ops claims and was not in the emergency-services archive.","streamIds":["emergency-services"],"roles":{"sales":"Sales framing: APCO/RapidSOS's October survey of 740 ECC professionals shows 42% already deploy or pilot AI while only about 8% completed any formal or informal AI risk assessment and mean high-stakes trust is 2.5/5, with accuracy the #1 barrier. For fire/EMS/911 buyers, the bounded offer is governance and evaluation help that matches what ECCs say they lack—not a claim that any product clears the 75% accuracy concern. Do not imply the survey endorses a vendor.","engineering":"Engineering framing: default architectures to transcription, translation, and QA assist with mandatory human final authority on dispatch recommendations; instrument accuracy and override rates because buyers cannot easily evaluate reliability pre-purchase (survey finding). Integrate CAD carefully—lack of CAD integration ranks among barriers. Treat policy/risk-assessment artifacts as deployment prerequisites.","delivery":"Delivery framing: every ECC AI pilot should include a documented risk assessment, named legal oversight, staff trust baseline (e.g., 1–5 scale), and accuracy test plan before expanding beyond assistive modes. Acceptance mirrors survey gaps: risk assessment on file, policy in progress or adopted, and measured accuracy—not deployment percentage alone."},"roleDetails":{"sales":{"customerProblem":"ECC leaders face pressure to adopt AI amid staffing strain but lack trust, risk assessments, and clarity on liability.","stakeholders":["911/ECC director","Fire/EMS chiefs sharing the center","County counsel","Telecommunicator workforce/union","Elected officials","APCO/NENA peers"],"discoveryQuestions":["Has a documented AI risk assessment been done?","Who owns legal oversight if AI contributes to a bad outcome?","What accuracy evidence exists on local call audio?","Is there a public notice plan for AI use?"],"valueHypothesis":"A governance-first engagement that closes the survey's risk-assessment and policy gaps can unlock safer pilots where agencies already want transcription/non-emergency automation.","potentialEngagement":"ECC AI readiness assessment mapped to the survey's risk, policy, and trust items, producing a board-ready gap report.","unsupportedClaims":["That 42% adoption proves any product is safe","Guaranteed accuracy improvements","That RapidSOS sponsorship implies product neutrality on all findings"]},"engineering":{"fit":"Fits assistive ECC AI (transcription, translation, QA, non-emergency automation) with human judgment retained on high-stakes decisions.","architectureIntegration":"AI services alongside CAD/call-handling with override, logging, and no silent automation of dispatch recommendations until accuracy gates pass.","prerequisites":["Documented risk assessment","CAD integration plan","CJIS/privacy review","Accuracy evaluation dataset"],"constraints":["Survey is perception/self-report not bench accuracy","Mixed agency types in sample","Vendor co-brand"],"security":"Address cyber, data security, and privacy risks respondents ranked highly; clarify vendor vs agency data ownership.","proofOfValue":["Pre/post trust score on Q29-like scale","Accuracy audit on held-out calls","Completion of formal risk assessment artifact"]},"delivery":{"work":"Stand up risk assessment, policy, oversight assignment, and accuracy pilot before scaling AI in the ECC.","dependencies":["Legal counsel","CAD vendor","QA/training unit","Labor engagement"],"ownership":"ECC director owns AI use policy; counsel owns liability posture; operations owns day-to-day overrides.","skillsAdoption":"Train PSTs on when to distrust AI outputs; supervisors audit AI-influenced calls.","governanceCheckpoints":["Risk assessment complete","Policy adopted or timed plan","Public communication decision","Accuracy gate before expanding use cases"],"acceptanceCriteria":["Written risk assessment on file","Named legal oversight role","Measured accuracy/override metrics meeting local bar"],"risks":["Deploying without assessment (majority posture in survey)","Liability ambiguity after an AI-touched incident","Staff distrust blocking adoption"]}},"retrievedAt":"2026-11-07T04:08:40Z","enrichedAt":"2026-11-07T04:12:46Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Translation and transcription top desired capabilities, suggesting accessibility gains for non-English callers, while over-reliance eroding skill is a top-rated risk—training and staffing models must preserve PST judgment.","procurementImplications":"RFPs should demand accuracy evidence, CAD integration, CJIS/privacy detail, and risk-assessment support—the items ECCs say are hardest to evaluate and most needed.","operatingModelImplications":"Operate AI as assistive under explicit human authority until trust and formal risk processes mature; publish community notices as centers move beyond pilots.","peerReview":"unknown","sourceVerification":{"openedUrl":"https://info.rapidsos.com/hubfs/RapidSOSxAPCO-International-State-of-AI-in-Public-Safety-Survey-2026.pdf","referenceExcerpt":"8% have completed a formal or informal AI risk assessment","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}