From the State Government edition of September 23, 2026
Oregon EO 26-26 directs third-party frontier AI safety reviews and kill-switch assessment for state procurement
Office of the Governor of Oregon (Tina Kotek) / State Chief Information Officer · State executive AI procurement and shared IT governance · Oregon, United States
- Publisher
- State of Oregon Executive Order No. 26-26 — Establishing Responsible Artificial Intelligence Procurement Standards for State Government
- Original publication
- 2026-09-23
- Source retrieved
- 2026-09-24
- Event date
- 2026-09-23
What happened
Why now: on 2026-09-23 Governor Kotek signed EO 26-26, stating Oregon policy to seek frontier AI models that have received independent third-party AI safety review. The State CIO must, within 90 days, submit an implementation proposal covering a frontier-model definition; standards for adequate third-party safety review; evaluation of a kill-switch requirement’s viability; methods to assess reviewer independence/qualifications/methodology/transparency; integration into procurement and technology-use practices; exceptions when review is unavailable; ongoing monitoring; transparency recommendations; and any needed statutory/administrative/budgetary changes. Order is effective immediately, remains until terminated, and must be reassessed every three months. Builds on EO 23-26 AI Advisory Council work and existing Responsible AI Usage Policy.
Why it matters
Direct state-government procurement control for frontier models—transferable to other states designing purchase gates for advanced GenAI without waiting for federal standards; explicitly uses contracting power rather than only agency-use policy.
Evidence and measured results
Primary: four-page official EO PDF OCR’d after HTTPS download. Corroborated by KVAL (updated 2026-09-23) and EIN/Oregon DEQ release summarizing the same directives. No measured pilot outcomes in the order itself.
Limitations and uncertainty
Standards and kill-switch requirement are not yet operational—only a 90-day proposal mandate. OCR of scanned PDF may contain minor transcription artifacts; prefer final oregon.gov posting when available. Political statements about federal leadership are not treated as empirical findings.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-23; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- States want frontier GenAI capabilities but lack agreed third-party safety-review and shutdown criteria in procurement.
- Stakeholders
- Governor’s office / SCIO
- Agency CIOs
- State procurement
- Cyber/privacy officers
- Budget staff
- Discovery questions
- Which current contracts would meet a draft frontier definition?
- What third-party evals do vendors already have?
- Who owns kill-switch testing if required?
- What exception path is needed for public-safety tools?
- Value hypothesis
- A bounded 90-day proposal-support engagement can translate EO 26-26 into actionable procurement criteria without promising certified safe models.
- Potential engagement
- 6–8 week frontier-procurement standards workshop and contract-clause drafting support for the SCIO proposal.
- Unsupported claims
- Guaranteed federal alignment
- Vendor self-attestation equals independent review
- Kill switch already proven in state production
Pre-sales engineering
Role takeaway
- Fit
- Fit for adding evaluation gates and contractual controls around frontier model APIs/platforms; unfit as a substitute for application-level safety testing.
- Architecture and integration
- Insert review artifacts into IT investment/procurement gates; tag frontier models in inventory; design shutdown/disable paths per hosting model.
- Prerequisites
- Model inventory with capability tags
- Procurement clause library
- Reviewer evaluation rubric
- Exception workflow
- Constraints
- Oregon executive-branch scope
- 90-day proposal clock
- Federal/state law still apply
- OCR/primary text may be refined on oregon.gov
- Security
- Independent review does not replace agency data-classification, logging, or access control for GenAI workloads.
- Proof of value
- Gap score of top 10 GenAI contracts vs draft criteria
- Tabletop kill-switch test for one hosted model
- Exception path exercise
- Quarterly reassessment checklist
Delivery
Role takeaway
- Work
- Facilitate SCIO proposal drafting, stakeholder interviews, and pilot clause insertion on a non-production or upcoming buy.
- Dependencies
- SCIO sponsorship
- DAS procurement counsel
- Agency application owners
- Existing Responsible AI Usage Policy owners
- Ownership
- State CIO with Governor approval for policy implementation.
- Skills and adoption
- Procurement officers need AI-eval literacy; engineers need shutdown runbooks.
- Governance checkpoints
- Draft frontier definition review
- Proposal submission day 90
- Governor approval gate
- Quarterly EO reassessment
- Acceptance criteria
- Written proposal covering all EO-required elements
- Published interim guidance for new frontier buys
- Documented exception criteria
- Monitoring plan named
- Risks
- Agencies buy ahead of standards
- Kill-switch language without test evidence
- Over-broad frontier definition blocking low-risk tools
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Expect procurement and enterprise architecture reviews to add third-party safety-review artifacts and possible kill-switch contractual clauses for frontier models; SCIO coordination across agencies; exceptions path for specialized/public-safety needs when review unavailable.
Governance
Who approves, reviews and stays accountable for outcomes?
Shifts frontier AI buying toward documented independent assessment and quarterly gubernatorial reassessment; proposal may recommend statute/admin changes. Existing Responsible AI Usage Policy remains baseline for generative/agentic use.
Security and privacy
What data, permissions and controls need testing?
Frames frontier AI risks to privacy, cybersecurity, public safety, and civil rights; third-party review and potential kill-switch aimed at containment/safety—not a substitute for agency data-handling controls already in Oregon policy.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
EO focuses on procurement safety of frontier models rather than workforce accessibility; implementation should still preserve accessibility requirements already embedded in Oregon IT procurement.
Procurement
What should contracts, pricing and exit terms secure?
Core impact is on purchase/contract criteria for frontier AI—third-party review evidence, possible kill-switch terms, exception handling, and transparency about state frontier-model use.
Operating model
Which teams own the service once it runs?
Creates a time-boxed SCIO-led standards build with agency cooperation duties and quarterly gubernatorial revisit—operating model is policy-then-implement, not an immediate ban.
Publication history
- 2026-09-23State Government · Issue 184 resources
Stable resource ID: oregon-eo-26-26-frontier-ai-procurement-standards-20260923