Oregon Requires Outside Safety Reviews for Frontier AI Used by State Agencies

The order took effect immediately, but officials still must define which models it covers and what counts as an adequate review.

By 2 min read
Oregon Requires Outside Safety Reviews for Frontier AI Used by State Agencies
Oregon Requires Outside Safety Reviews for Frontier AI Used by State Agencies

Listen to this story

The audio brief

About 1:25
0:001:25
Read transcript
Oregon has made independent, third-party safety review a condition for the most advanced AI models its executive agencies buy or use. Governor Tina Kotek’s order took effect immediately. But the practical test is still unwritten: the state has not decided which systems count as frontier AI, or what makes a review adequate. The requirement applies to Oregon executive agencies’ purchases and use of AI. It does not require reviews for companies selling those models to other customers. The order builds on existing state rules for managing AI risks, including human accountability and oversight. Kotek said states should act while broader safeguards remain insufficient. Terrence Woods, Oregon’s chief information officer, has 90 days to propose a definition of frontier AI, standards for outside reviews, and guidance on how agencies should use those reviews in procurement, contracts, and day-to-day technology decisions. That is a deadline for a proposal—not for adopting the standards. The state has set no date for putting the detailed criteria in place. Woods must also assess whether requiring a kill switch is viable. The order does not require one; an emergency shutoff is still under consideration. Meanwhile, the order stays active until Kotek terminates it, and she will reassess its necessity every three months. The key constraint is that agencies now face the review condition, while the state is still working out exactly how to apply it.

Story brief

3 key points

Oregon’s new purchasing rule could make independent safety review a prerequisite for frontier AI vendors seeking executive-agency business, but its practical reach depends on standards the state has not yet written. Chief Information Officer Terrence Woods has 90 days to propose what qualifies as frontier AI and an adequate review, plus how agencies should apply reviews in procurement and use. There is no deadline...

  1. 01

    Executive Order 26-26, issued September 23, applies to Oregon executive agencies—not AI companies’ sales to other customers.

  2. 02

    Woods must also assess whether requiring a kill switch is viable; the order does not currently mandate one.

  3. 03

    The order remains in effect until Governor Tina Kotek terminates it. Oregon has not set a date for adopting the detailed criteria.

Oregon has put a safety condition on the most advanced AI models its executive agencies buy or use: they must receive independent, third-party review. Governor Tina Kotek’s order is already in effect, but the state still has to define which models qualify and what counts as an adequate review.

A new condition for state purchases

Kotek issued Executive Order 26-26 on September 23. It says Oregon will buy, contract for and use frontier AI models that have received independent safety review, consistent with state and federal law. The order governs state executive agencies’ purchasing and use; it does not impose a general review requirement on AI companies’ sales to other customers.

The step builds on statewide rules requiring executive agencies to manage AI risks and maintain human accountability, transparency and oversight. Kotek argued that states must act while federal safeguards remain insufficient. She said she would use her executive authority until Oregon lawmakers and Congress pass comprehensive standards for AI companies.

It's encouraging to hear AI companies are beginning to come forward with proposals to address safety concerns, but we cannot simply wait for the private sector to self-regulate a technology that poses such an existential and evolving risk.

Oregon Governor Tina Kotek, in a statement reported by KTVZ

Woods must turn the condition into a test

State Chief Information Officer Terrence Woods must develop criteria for an adequate outside safety review and a definition of frontier AI suited to state purchasing and use. He must also work out how agencies should incorporate reviews into procurement, contracts and technology-use practices. Those decisions will determine how officials apply the order to particular models and purchases.

Another assignment is to assess whether a kill-switch requirement for frontier AI models would be viable. An emergency shutoff is an option under study, not a safeguard the order has already required. That leaves a separate choice alongside the order’s condition for outside review.

A deadline for the proposal, not for adoption

Woods has 90 days to submit an implementation proposal to Kotek. Hope Hiebert, communications director at Oregon Enterprise Information Services, said it will address the criteria for adequate independent review. Oregon has no separate timeline yet for adopting any resulting standards, she said. The proposal therefore has a deadline, but the state has not set one for putting its detailed review criteria in place.

The order remains active until Kotek terminates it. She will reassess its necessity every three months, giving the policy a recurring review even as agencies await the standards for applying it.

Sources

  1. govtech.comOregon Is Latest State to Move on Frontier AI Regulation
  2. ktvz.comGov. Tina Kotek orders new AI safety standards for Oregon state agencies

Loading discussion...

YOUR READING SPACE

Notifications

Oregon Requires Outside Safety Reviews for Frontier AI Used by State Agencies | Superpower Daily