{"resourceId":"nature-disasteller-lvlm-disaster-response-20260110","versions":[{"version":"external-ef77c92f8ecde753fdccacac354ff4063d85943234f12e7a436d90281341b0f2","resource":{"id":"nature-disasteller-lvlm-disaster-response-20260110","title":"DisasTeller multi-LVLM disaster agents: early grading errors propagate; HITL audit required","organization":"Nature Communications (University of Sydney et al.)","sector":"Academic research / Emergency management","geography":"International case studies (earthquake/flood/bushfire); transferable methods caution for U.S. EM","publishedAt":"January 10, 2026 (VoR February 9, 2026)","publicationDate":"2026-01-10","eventDate":null,"sourceName":"Nature Communications","sourceLabel":"Nature Communications article","sourceUrl":"https://www.nature.com/articles/s41467-025-68216-z","evidenceClass":"academic-research","outcomeClass":"cautionary","topics":["developers-agents","operating-model","governance-procurement","infrastructure"],"finding":"DisasTeller coordinates four LVLM agents (GPT-4o vs Gemma3-27B) for post-disaster assessment, alerts, emergency services, and assignment/reconstruction. GPT-4o intermediate grading averaged 7.3/10 vs 4.6/10 for Gemma3; map annotation 6.0 vs 5.0. Authors show early-stage agent errors can propagate to allocations and plans, and argue intermediate artifacts must enable human audit/intervention. Framework is decision support, not replacement of authority.","sledRelevance":"Evidence that multi-agent LVLM ‘EOC copilots’ can accelerate drafting yet still need staged HITL because grading mistakes cascade—material for emergency management AI governance as agentic tools are pitched to EOCs.","evidence":"Peer-reviewed open-access study with repeated runs, human+EvaluatorGPT scoring, and explicit limitation analysis. Case studies use public crisis imagery; allocation numbers are illustrative, not optimized OR solutions.","architectureImplications":"Multi-agent orchestration with tools (web search, file read, map annotation) needs intermediate checkpoints and GIS grounding; open weights aid inspectability vs closed GPT.","governanceImplications":"Mandate human authority, logged intermediate grades, and prohibition on unsupervised public alerts from early agents.","securityPrivacyImplications":"Web-search grounding and image pipelines create exfiltration and misinformation risks; prefer curated institutional knowledge bases over open web in production EOCs.","caveats":"Published earlier in 2026 but not archived; newly relevant alongside SAFIRE as multimodal/agentic disaster AI moves from demos to procurement. Not a U.S. operational deployment.","streamIds":["emergency-services"],"roles":{"sales":"Sell DisasTeller-class systems as timed drafting aids with mandatory human gates—not autonomous EOC command. Use GPT vs Gemma score gaps to justify evaluation harnesses and open-weight options for inspectability.","engineering":"Architecture needs agent memory, tool sandboxing, intermediate map/grade artifacts, and stop points before public/assignment outputs. GIS integration is a stated missing piece for safe spatial advice.","delivery":"Tabletop the agent chain with real incident imagery, measure error propagation, train officers to override early grades, and keep final resource orders human-signed."},"roleDetails":{"sales":{"customerProblem":"EOCs want faster damage reports and allocation drafts without surrendering life-safety authority to hallucinating agents.","stakeholders":["Emergency managers","Situation unit leads","GIS/IT","Elected oversight"],"discoveryQuestions":["Which outputs may go public automatically?","Who signs resource orders?","What institutional docs replace open web?","Open-weight requirement?"],"valueHypothesis":"Multi-LVLM drafting can compress hours of coordination if every stage is human-auditable and early errors are catchable.","potentialEngagement":"Tabletop pilot on historical local disaster imagery with scored HITL checkpoints.","unsupportedClaims":["Minutes-scale end-to-end replaces human teams safely","Allocation counts are optimal","Closed GPT quality is guaranteed over time"]},"engineering":{"fit":"Crew-style multi-agent LVLM with tools for image grade, alert draft, and allocation narrative.","architectureIntegration":"Image in → expert agent grade → alerts/emergency/assignment agents → human approve; persist intermediates.","prerequisites":["EMS-98 or local damage rubric","Sandboxed tools","Human approval API","Optional GIS layers"],"constraints":["Error propagation","Non-deterministic outputs","Weak spatial safety reasoning"],"security":"Limit web tool; prefer private knowledge; redact sensitive imagery sharing.","proofOfValue":["Human catches injected early errors","Grade agreement vs SME panel","Time-to-draft with quality floor"]},"delivery":{"work":"Configure agents, rubric, approval UX, and after-action on propagation failures.","dependencies":["SME graders","Secure LLM hosting","GIS team","Public information officer rules"],"ownership":"Emergency management director retains decision authority.","skillsAdoption":"Officers practice challenging fluent agent drafts; avoid moral deskilling.","governanceCheckpoints":["No unsupervised public alert","Stage-gate after grading","Model version pin + eval"],"acceptanceCriteria":["All intermediates stored","Human signature on assignments","Documented override rate"],"risks":["Cascading wrong grades","Public misinformation","Over-reliance under fatigue"]}},"retrievedAt":"2026-09-24T22:02:40-05:00","enrichedAt":"2026-09-24T22:02:40-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Speedy agent drafts can widen EM capacity if humans stay skilled reviewers; risk of deskilling if overrides are rare.","procurementImplications":"Require intermediate audit trails, HITL gates, and model-card scorecards (closed vs open) in EOC AI buys.","operatingModelImplications":"Reframes AI as parallel drafting lane beside human ICS sections, not a replacement command chain.","sourceVerification":{"openedUrl":"https://www.nature.com/articles/s41467-025-68216-z","referenceExcerpt":"errors in early-stage agents (e.g., disaster grading) can propagate to downstream outputs","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}