Strategic PartnersIssue 124 records
Microsoft
Four items. Ohio State and Microsoft announced a five-year partnership on Nov. 10 to move about 70-80% of the medical center's storage and compute to Azure, build a Fabric data foundation and expand Dragon Copilot note drafting (vendor claims; one self-reported facility metric, no financial terms). Microsoft's Nov. 10 security release included a critical Azure OpenAI flaw, CVE-2026-119990 (CVSS 9.8, no authentication required); ZDI notes the guidance changed on release day to 'no customer action needed', and it is not reported as exploited. Portland's Microsoft 365 Copilot privacy impact assessment, which lists 48 risks and excludes police use, reaches its stated validity end today, and Australia's DTA guidance (via IDM) warns that Copilot outputs may not be reproducible for records purposes. No new U.S. public-safety, HHS, transportation/utility or K-12 Microsoft AI evidence surfaced.
- Evidence records
- 4
- Cross-source patterns
- 1
- Evidence classes
- 1 independent reporting1 independent research1 government evaluation1 standards or public-body guidance
- Outcomes
- 3 cautionary1 emerging
- Source freshness
- 3 new this fortnight1 older, newly relevant
- Research completed
- 2026-11-11
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Government Copilot reviews place records capture and periodic reassessment on the deploying agency, not the platform
Portland's privacy assessment says Copilot outputs and interaction logs can become public records and sets a fixed validity date (Nov. 10, 2026), while Australia's DTA guidance, as reported by IDM, warns that outputs may not be reproducible and that audit capture depends on configuration, so relied-upon outputs must be saved under agency rules. Both treat Copilot data-source configuration as exposing existing permission and classification weaknesses. Neither reports measured incidents; together they suggest SLED Copilot governance needs scheduled re-review as features such as Graph grounding and agents expand.
Operating questionWhen does your Copilot privacy or records review expire, and does it cover how relied-upon outputs are captured as records?
Supporting evidenceCity of Portland, Oregon (Smart City PDX privacy program; Bureau of Technology Services)Australian Digital Transformation Agency (reported by Information & Data Manager)
Evidence ledger
Full record · every source keeps its link and limitations
Ohio State and Microsoft sign five-year partnership: 70-80% of storage and compute to Azure, Dragon Copilot notes, Fabric data foundation
On Nov. 10 Ohio State and Microsoft announced a five-year partnership covering a smart-hospital model, moving about 70% to 80% of workloads to Azure, a Microsoft Fabric/OneLake data foundation, and AI support for clinicians. The Lantern reports the medical center's chief health informatics officer said early smart-hospital work improved temperature control from 88% to 98%. Microsoft's same-day press release (vendor claim) says clinicians already use Dragon Copilot ambient note drafting with clinician review, the data platform's high-level design is complete, the EHR will be backed up on Azure for continuity, and tumor-board AI is only being explored. The Next Web separately reported no financial terms were disclosed.
Why it matters, evidence and limitations
- Why it matters
- A large U.S. public university's academic medical center is committing most of its storage and compute to Azure and pairing it with ambient clinical AI and Microsoft Learn certification. That makes it a reference account for public universities with health systems and for state workforce programmes. Interpretation: most components are plans or early pilots, so it is a signal of direction, not evidence of outcomes.
- Evidence and measured results
- Student-newspaper reporting from the announcement event with quotes from Ohio State and Microsoft executives, one institution-reported metric (temperature control 88% to 98%), and the vendor press release. No contract value, baseline method, clinician-time data or note-accuracy data.
- Limitations and uncertainty
- Announcement-day reporting; the 88%-to-98% figure is self-reported without method or baseline period. Most work is planned (data platform at design stage, tumor-board AI exploratory). No financial terms. TNW notes AI scribes can make serious errors; no accuracy data from Ohio State is provided.
November Patch Tuesday: critical Azure OpenAI flaw CVE-2026-119990 (CVSS 9.8, no authentication), with conflicting guidance on customer action
Microsoft's Nov. 10 release included CVE-2026-119990, an Azure OpenAI elevation-of-privilege flaw rated Critical, CVSS 9.8. ZDI says the advisory FAQ describes an unauthenticated attacker sending a crafted internet request to an affected Azure OpenAI service, with impact on orchestration information, workflow changes or service disruption. ZDI notes the bulletin first said 'Customer Action Required: Yes' without steps, then Microsoft revised it to say no customer action is needed. NVD lists it as CWE-306 (missing authentication), published Nov. 10, awaiting analysis; CSO Online reports the service is patched by Microsoft. ZDI lists it as neither public nor exploited. Azure OpenAI was one of four Critical Azure and identity bugs.
Why it matters, evidence and limitations
- Why it matters
- State agencies, cities and universities running chatbots, document processing or agents on Azure OpenAI depend on Microsoft's server-side fix, so the work for SLED teams is verification and documentation, not patching. Interpretation: a critical no-auth flaw in a managed AI service is a reminder to include AI platform services in vulnerability tracking and vendor-risk reviews.
- Evidence and measured results
- Independent security-research analysis of Microsoft's advisory data, corroborated by the NVD record (CVSS 3.1 vector AV:N/AC:L/PR:N/UI:N, 9.8) and CSO Online reporting. No exploitation data; affected-build detail and exposure window not confirmed from Microsoft's advisory page, which was not opened.
- Limitations and uncertainty
- Microsoft's own advisory (msrc.microsoft.com) was not opened directly; details are from ZDI, NVD and CSO. No confirmed exposure window, affected configurations or exploitation evidence. Guidance changed on release day and may change again.
City of Portland's Microsoft 365 Copilot privacy impact assessment (48 risks, police excluded) reaches its stated validity end on Nov. 10, 2026
Portland's 79-page privacy impact and risk analysis of Microsoft 365 Copilot for city staff identified 48 risks across six areas; the worst case after mitigation was weighted Medium-high. It excludes Portland Police Bureau use. Highlighted risks include unintentional sharing of sensitive data in prompts (Medium), oversurveillance of staff through productivity data (Medium-high) and aggregation of information about individuals. It says Copilot outputs and interaction logs can become public records subject to disclosure. Copilot Chat initially had no Microsoft Graph access, though the assessment covers Graph-integrated risks. BTS adopted Copilot as Standard Technology, was approving a use policy and required training. The document states it is valid until Nov. 10, 2026.
Why it matters, evidence and limitations
- Why it matters
- Why now: the assessment's stated validity ends on this edition date, and no renewed assessment was found in search. U.S. cities and counties can use it as a model risk register for Copilot and as a reminder that privacy reviews expire while licensing and features (Graph grounding, agents) keep expanding.
- Evidence and measured results
- Government-authored privacy risk analysis using the Smart City PDX impact-likelihood matrix; qualitative ratings, no usage data, incident counts or outcome measures.
- Limitations and uncertainty
- Released December 2025, so it predates recent Copilot changes (Copilot Chat fair-use limits, agents, new models). Ratings are expert judgments, not measured incidents. Whether Portland has renewed or extended it is unknown; absence of a renewal in search results is not proof none exists.
Australia's DTA AI Reference Architecture flags a Microsoft 365 Copilot records gap: outputs may not be reproducible and must be captured by agencies
IDM reports that the DTA's AI Reference Architecture (AIRA v1.0, published Oct. 30) and its Companion Guide use Microsoft 365 Copilot as a scenario. The guide notes audit logs may record prompts, outputs or referenced files depending on configuration and tooling, cautions that outputs are not necessarily reproducible from the same prompt, and says outputs relied on for business purposes should be captured under agency record-keeping rules. It says configuring Copilot's data sources may expose existing weaknesses in permissions, classification, information architecture and records management. AIRA is voluntary and scales traceability from source attribution for productivity assistants to decision reconstruction for consequential uses.
Why it matters, evidence and limitations
- Why it matters
- U.S. state and local records laws impose similar duties to retain records of official business. The DTA's point that Copilot outputs cannot be regenerated on demand supports capturing relied-upon outputs at the time of use rather than counting on logs. Interpretation: applies to SLED records managers setting Copilot retention policy.
- Evidence and measured results
- Trade-press summary of national government guidance with direct quotes; the DTA's original PDF and companion-guide page could not be retrieved (connection errors), so details rely on IDM's reporting.
- Limitations and uncertainty
- Australian federal guidance under the Archives Act and PSPF; U.S. public-records and retention schedules differ by state. Published Nov. 6, before the last completed microsoft run, but not previously archived; included for its direct relevance to the Portland assessment in this edition. Original DTA documents not opened; findings rely on IDM.
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.
- Independent research
- Research conducted outside the implementing organization.
- Government evaluation
- A public body’s measured evaluation or documented pilot.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.