State & Local GovernmentIssue 184 records
Local Government
Four new-to-archive sources: Colorado Springs went live Sept. 23 with Axon S.A.R.A.H. AI answering non-emergency calls after betas where roughly half finished without a dispatcher, at $170,000/year with human review; Raleigh told Smart Cities Connect it will not keep Daisy after missing a 90% correct-answer bar despite cutting factual errors to ~5%; Sheffield, Ala. unanimously bought SuperGrok for $3,000 with a human-send gate for project coordination; and UNC Coates’ Canons’ Sept. 9 14-point AI use-policy checklist is newly salient beside those same-day go-live and stop decisions.
- Evidence records
- 4
- Cross-source patterns
- 2
- Evidence classes
- 3 independent reporting1 standards or public-body guidance
- Outcomes
- 3 emerging1 cautionary
- Source freshness
- 3 new this fortnight1 recent
- Research completed
- 2026-09-24
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Publish accuracy gates—and honor stop decisions—before six-figure resident AI spend
Raleigh pre-committed to a 90% correct-answer bar, found early ~30% accuracy and lingering empty answers, and did not keep Daisy. Colorado Springs went live only after four betas and kept humans reviewing every AI-produced non-emergency call. Together they argue cities should write kill criteria and human-review SLAs before production spend.
Operating questionWhat correct-answer, non-answer, and human-review thresholds must be met before resident-facing AI stays in production?
Supporting evidenceCity of RaleighColorado Springs Police Department / City of Colorado Springs
Even cheap staff AI needs human-send controls and written use policies
Sheffield’s $3,000 SuperGrok purchase emphasizes no send/post until a human acts, while UNC Coates’ Canons urges IT approval, confidential-data bans, fact-checking, and annual policy review amid shadow AI. Small-dollar tools still need the same accountability scaffolding as larger chatbots.
Operating questionBefore enabling staff AI seats, are human-gated outbound actions, IT approval, and fact-check duties written and trainable?
Supporting evidenceCity of SheffieldUNC School of Government (Coates’ Canons)
Evidence ledger
Full record · every source keeps its link and limitations
Colorado Springs goes live with Axon S.A.R.A.H. AI for non-emergency calls
On Sept. 23, 2026 The Gazette reported that an AI chatbot named S.A.R.A.H. (Service Assessment and Response Assistance Helper), powered by Axon’s Prepared AI, began answering all non-emergency calls to the Colorado Springs communications center that Tuesday. Deputy Police Chief John Koch said the goal is shorter waits so dispatchers can prioritize 911: emergency answer times average 11–13 seconds while non-emergency waits usually exceed 30 seconds; with AI there is now almost no wait for non-emergency calls. Across four beta tests, CSPD found roughly half of non-emergency calls could be completed without a human dispatcher; the bot asks questions to decide whether a human call taker is needed. Koch said humans review each AI-produced call and callers can request a human. Contract cost for Prepared AI by Axon is $170,000 annually per a CSPD spokesperson. CSPD already uses other Axon tools (drones, body-cam report drafting, LPR). Koch discussed possible future developmental partnering including remote dispatcher answering.
Why it matters, evidence and limitations
- Why it matters
- Same-day U.S. city go-live of vendor non-emergency voice AI with published beta completion share, wait-time claims, annual contract cost, and mandatory human review—directly relevant to municipal resident-access and communications capacity.
- Evidence and measured results
- Gazette Sept. 23, 2026: S.A.R.A.H. live Tuesday for all non-emergency calls; Axon Prepared AI; $170,000 annually; four beta tests ~half completed without human dispatcher; human review of each AI-produced call; caller can request human; emergency 11–13s vs non-emergency >30s baseline; almost no wait after go-live per police.
- Limitations and uncertainty
- Local press quoting CSPD; beta “roughly half” and wait-time claims are department-reported, not an independent audit; Axon ecosystem concentration noted in article but not evaluated.
Raleigh ends Daisy resident AI pilot after missing 90% correct-answer bar
On Sept. 23, 2026 StateTech reported that Raleigh tested Daisy, an AI customer-service assistant with a humanlike avatar, from April through July and ultimately decided not to keep it. Officials set a firm bar: at least 90% correct answers before spending hundreds of thousands on broader deployment. Internal testing with 25 employees showed about 30% accuracy and ~4/10 satisfaction. An outdated strategic plan naming a prior mayor/council polluted answers (“garbage in, garbage out”). By pilot end only about 5% of answers were incorrect, but many replies were non-answers lacking information, so the 90% goal was still missed. Focus groups split on whether a human face helped; residents preferred people for emotional/complex needs and were more open to AI for queue-jumping simple questions or after-hours help. Takeaway emphasized staff-facing AI for tedious work and needs-based placement rather than slapping AI on everything. Cited 2025 PayIt index: 83% of governments implementing AI vs 50% consumer comfort.
Why it matters, evidence and limitations
- Why it matters
- Same-day cautionary measured pilot: explicit accuracy gate, data-hygiene failure mode, avatar UX split, and decision not to procure—contrasts with go-live announcements peers may over-weight.
- Evidence and measured results
- StateTech Sept. 23, 2026: Daisy April–July pilot; not retained; 90% correct bar; internal ~30% accuracy / ~4/10 satisfaction; outdated strategic plan caused errors; end-state ~5% incorrect but too many “don’t have information” replies; avatar face controversial; complex/emotional questions prefer humans; staff-facing AI may offer more value.
- Limitations and uncertainty
- Conference reporting of city officials’ briefing; exact vendor and public sample sizes beyond the 25-employee internal test are limited in the article; PayIt comfort statistic is secondary citation.
Sheffield, Ala. council buys SuperGrok for $3,000 project coordination assist
On Sept. 23, 2026 WAFF reported Sheffield, Alabama Mayor Bryan Van Devender said the city now employs the SuperGrok platform to help organize city work. City Council unanimously approved buying the program’s highest capacity for $3,000. The mayor frames use for coordinating paperwork, staffing locations, and stakeholders across five concurrent grant-funded projects plus other work—not replacing humans answering phones, contractors, or council fiscal control. He says any city employee may use it for city purposes, rollout is forthcoming, and use is optional. The program is described as programmed not to send or post anything until a human has reviewed and the human sends. No productivity metrics or security review details appear in the article.
Why it matters, evidence and limitations
- Why it matters
- Same-day small-government capacity signal: low-dollar generative assistant for grant/project coordination with explicit human-send control—useful peer contrast to six-figure voice AI and chatbot procurements.
- Evidence and measured results
- WAFF Sept. 23, 2026: SuperGrok; unanimous council; $3,000 highest capacity; project/grant paperwork coordination; mayor says not replacing humans/phone/council; employee optional access; will not send/post until human review and human send.
- Limitations and uncertainty
- Local TV paraphrase of mayor remarks; no independent security assessment, vendor contract text, or measured time savings in the article.
UNC Coates’ Canons issues 14 recommendations for local-government AI use policies
On Sept. 9, 2026 UNC School of Government’s Coates’ Canons published Kristi Nickodem’s updated 14 recommendations for local-government AI use policies. The post argues many units still lack guardrails despite shadow AI: it cites a 2025 KPMG/University of Melbourne survey that 58% of employees intentionally use AI regularly and 70% of those use free public tools. Recommendations include: don’t copy-paste policies; define covered technology; require IT review for new software/subscriptions; set agentic-AI boundaries; prohibit improper confidential disclosures; require fact-checking and accountability for outcomes (noting LLM hallucinations); consider department-specific rules; remind staff of public-records duties; address transcription/notetaking/recording; state prohibited uses; calibrate transparency; involve the right people; review policies at least annually; and test whether employees can understand/apply the policy. Why-now for the 2026-09-23 edition: same-day Raleigh stop decision and Colorado Springs go-live make written fact-check, IT-approval, and prohibited-use rules immediately operational for peers.
Why it matters, evidence and limitations
- Why it matters
- Actionable local-gov policy checklist newly salient beside same-day production go-live and cautionary non-retention stories; fills the controls gap when cities buy voice AI or $3k LLMs.
- Evidence and measured results
- Coates’ Canons Sept. 9, 2026: 14 recommendations listed; shadow-AI rationale with 58%/70% survey cite; fact-checking/hallucinations; IT approval; confidential disclosures; agentic boundaries; annual review; employee understandability test.
- Limitations and uncertainty
- Guidance blog, not an empirical evaluation of any city’s outcomes; survey statistics are secondary citations from 2025; older than last edition but included with explicit why-now beside 2026-09-23 deployments.
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.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.