Oregon AI Ethics Rules for Lawyers

The official instruments

A numbered formal opinion that leads with competence

Oregon answered the AI question with a true formal opinion. Formal Opinion 2025-205, “Artificial Intelligence Tools,” was approved by the Oregon State Bar Board of Governors in February 2025, and the label matters: this is a numbered formal opinion, the most authoritative kind of bar instrument, not informal guidance, a bar-counsel article, or a working-group report. That puts Oregon among the states that have committed to formal-opinion treatment of AI rather than the lighter-touch website notices some jurisdictions chose. The opinion carries a February 2025 month date; the published date is the month, so this tracker cites February 2025 and does not assert a specific day.

What gives Formal Opinion 2025-205 its character is the order of its priorities. It leads with competence, framed as an ongoing obligation rather than a one-time hurdle. A lawyer must maintain technology competence as AI tools evolve, which means the duty to understand a tool does not end once the lawyer has used it once; it continues as the tool changes and as new tools appear. That ongoing framing connects directly to the opinion’s most practical instruction, which is to vet a tool before relying on it. A lawyer cannot run the confidentiality analysis the opinion requires, or judge whether an output needs heavier review, without first understanding how the specific tool actually works.

The opinion permits AI use, but it surrounds that permission with qualifications. Beyond competence, it addresses client communication and possible disclosure depending on case-specific factors, declining to impose a blanket disclosure rule while leaving room for circumstances that require it. On fees, it provides that a lawyer should not charge a client to learn AI absent a client request and a clear agreement, treating learning time as overhead in line with ABA Formal Opinion 512. And it imposes supervision and training duties, the firm-level obligation that runs through every state instrument.

The confidentiality and verification core

Two requirements form the operational heart of Formal Opinion 2025-205. The first is a confidentiality analysis built around the difference between open and closed AI models. The opinion does not treat all AI tools alike for confidentiality purposes; it distinguishes open models, where inputs may be exposed or used to train the system, from closed ones with tighter data handling, and it requires informed consent before a lawyer puts client information into an open model. That open-versus-closed distinction is a more precise instruction than a generic “protect confidences” rule, and it forces a lawyer to know, for each tool, which category it falls into, which is exactly why the competence duty comes first.

The second requirement is mandatory accuracy review. Formal Opinion 2025-205 requires review of any output that contains case-specific facts, citations, quotations, or conclusions. That list is worth reading carefully, because it is specific: it is not only fabricated case citations that trigger the duty, but quotations and conclusions as well. The opinion treats AI output as a draft to be independently confirmed before it is relied upon or filed, which is the anti-fabrication principle that defines the entire field. Taken together, the confidentiality analysis and the accuracy-review duty mean an Oregon lawyer has to make two judgments about every AI tool and every AI output: is this tool safe for this information, and has this output been independently verified. The policy template builds both judgments into a repeatable workflow, and the comparison table places Oregon’s consent posture among the states. The tracker lists the opinion with its honest type and date.

What Oregon lawyers should do now

Operationalize the two core duties. For confidentiality, maintain an approved-tools register that records, for each tool, whether it is an open or closed model and what its data-handling terms are, because Formal Opinion 2025-205 turns the consent question on that classification; obtain informed consent before any client information goes into an open model. For accuracy, adopt a verification workflow that treats every output containing case-specific facts, citations, quotations, or conclusions as a draft requiring independent confirmation, matching the opinion’s specific trigger list. On fees, do not bill clients to learn AI absent a request and a clear agreement, and document any agreed AI charges. On supervision, maintain firm training and a written, acknowledged policy. Because the opinion frames competence as ongoing, the register and the workflow should be reviewed as tools change, not set once. The changelog records any movement on the Oregon entry, and the disclosure comparison shows how Oregon’s case-specific communication posture compares.

One layer the opinion does not address is the courts. Oregon has no verified statewide court rule on AI in filings, so the obligations that exist at the courthouse are judge-by-judge standing orders rather than a uniform rule. A lawyer filing in Oregon state or federal court should check the assigned judge’s standing orders before each filing, because an individual judge may require a disclosure or certification that the bar opinion does not. The court orders guide covers that per-filing check.