Oregon Formal Opinion 2026-208: AI Intake for Attorneys

If you're a solo or small-firm attorney in Bend, you've probably noticed the pattern. The calls come in at 7:42 a.m. when you're walking into court, at 6:30 p.m. when you're already home, and on Saturday morning when somebody's spouse just got served. You miss them. The next firm on Google picks up. The case goes there.

AI intake tools — voice receptionists and website chatbots that can answer the phone twenty-four hours a day, run a basic conflict check, and book a consultation — are the obvious fix. Until February 2026, Oregon attorneys had to use them without specific guidance from the bar. That changed when the Oregon State Bar Board of Governors approved Formal Opinion 2026-208, titled Chatbots and AI Agent Communications.

The opinion's conclusion is one line: "Yes, qualified."

This piece is about the qualifiers — what they actually mean, and what to do about them before you sign up for any AI intake tool.

The TL;DR for busy attorneys

• The Oregon State Bar approved AI chatbots and voice agents for client intake in February 2026.
• The opinion explicitly covers both text and voice AI — most commentary has missed this.
• Five Oregon Rules of Professional Conduct are implicated: RPC 1.1 (Competence), RPC 1.18 (Duties to Prospective Clients), RPC 5.3 (Responsibilities Regarding Nonlawyer Assistance), RPC 7.1 (False or Misleading Communications), and RPC 7.3(a) (Solicitation).
• The two biggest practical risks are (1) accidentally forming an attorney-client relationship through chatbot conduct and (2) collecting "disqualifying information" from a prospective client that conflicts you out of a different representation.
• You are supervising a vendor under RPC 5.3 — not just buying software.
• A human (you or a trained staff member) must be able to review and modify what the chatbot communicates.

What the opinion actually says — and why "chatbot" includes your AI receptionist

The opinion defines a chatbot as a software application designed to simulate conversation with human users, typically through text or voice interactions. That phrase matters. Every existing commentary on 2026-208 has treated it as guidance about website widgets — the little chat box that pops up in the corner. It is not. The Oregon State Bar's definition of "chatbot" expressly captures voice AI receptionists: software that answers your phone, books your appointments, and follows up by text.

If you've been evaluating products like Smith.ai, Ruby, Answering Legal, or any newer AI-first voice receptionist, you're inside the scope of this opinion. The fact that the receptionist speaks instead of types does not change the analysis.

The opinion also draws a sharp distinction between two technical categories of chatbot, and this distinction matters more than most lawyers will assume.

Rule-based chatbots follow predefined scripts. They have conversational ability but low autonomy. They do roughly what they were programmed to do, and not much more.

AI-agent chatbots are different. They may run on top of a large language model and operate as autonomous agents — software that can take actions, make decisions, and even spawn sub-agents to accomplish related tasks. The opinion gives a vivid example: a lawyer who gives an agentic AI the goal "engage a person to become a client" may end up with sub-agents that gather intake information, draft an engagement letter, and integrate that information into the firm's systems — all without the lawyer's intervening input.

That example is the heart of the opinion's concern. It is also the heart of yours.

The five Oregon Rules of Professional Conduct that apply

RPC 1.1 — Competence

The opinion is direct: an attorney using a chatbot must have taken reasonable steps to become competent in the technology. Competence here is not abstract. It includes understanding how the chatbot stores information, how it responds to prompts, whether it learns from user inputs, and what other vendors sit underneath your vendor's software stack.

The opinion specifically notes that this evaluation is not a one-time exercise. AI products evolve constantly. A chatbot that was compliant in March may behave differently in September because the underlying model was updated. Your due diligence is ongoing.

And it's substantial. The opinion notes that adequately understanding a chatbot may require lawyers to conduct research, attend training, or hire a professional to assist them. For a solo attorney, the practical translation is: budget time, not just money.

RPC 1.18 — Duties to prospective clients (the one that will bite solos)

This is the rule most attorneys underestimate, and it is the most likely to cause real harm.

Under RPC 1.18(b), you owe duties of confidentiality to people who consult with you about possibly retaining you, even if no attorney-client relationship ever forms. Under RPC 1.18(c), if you receive "disqualifying information" from a prospective client, you may be barred from representing a different client whose interests are materially adverse to that prospective client in the same or a substantially related matter.

Now imagine that disqualifying information was collected by a chatbot at 11 p.m. on a Saturday, while you were asleep.

This is not a hypothetical. A well-designed AI intake tool will ask probing questions to qualify a lead. The more thorough the intake, the more likely the conversation produces information that imputes a conflict to your firm. If a prospective client tells your chatbot the names of the parties, the nature of the dispute, and the strategic posture they're considering — and then their opposing party calls you the following Tuesday — you may already be disqualified, and you would never have known.

The opinion's practical answer is to enter prospective clients into your conflict management system promptly. The deeper answer is to design your intake to collect the minimum information necessary on the front end, and defer detailed intake to a human conversation that happens after a conflict check.

RPC 5.3 — Supervision of nonlawyer assistance

This is the rule that reframes how you should think about your AI vendor.

The opinion treats your chatbot provider as a third-party nonlawyer service provider, and you as the lawyer supervising that provider. The supervision is real, not nominal. The opinion explicitly references reference checks, vendor credentials, security policies, the vendor's hiring practices, and written confidentiality agreements.

This is a meaningful shift in posture. You are not a buyer of software. You are a supervising lawyer responsible for the conduct of a nonlawyer assistant — and your assistant happens to be a piece of software hosted by a third party that in turn relies on other third parties.

The opinion does include one quiet but important practical accommodation. In footnote 10, it notes that the task of reviewing chatbot communications, with appropriate training, may be delegated to staff. For solos and small firms without time to personally read every transcript, this is the permission slip that makes the whole arrangement workable.

RPC 7.1 — False or misleading communications

Your chatbot cannot misrepresent your expertise, your past results, or your services. The opinion observes — citing Florida's Ethics Opinion 24-1 and the New York City Bar's Formal Opinion 2024-5 — that a chatbot that fails to identify itself as a chatbot is, by itself, potentially a misleading communication.

The practical implication is straightforward: at the start of every interaction, your AI must identify itself as an AI. Not after the caller asks. Not in a footer disclaimer they won't read. At the start.

The opinion also reminds you that any communication the chatbot makes is one you are responsible for under the advertising rules. If the chatbot tells a prospective client you have thirty years of experience and you have eight, that is your problem, not the vendor's.

RPC 7.3(a) — Solicitation of vulnerable persons

A human intake person can tell when a caller is in crisis. They hear the trembling voice, the long pauses, the disorientation. They know when to stop the intake and ask whether the caller is safe.

A chatbot may not.

RPC 7.3(a) prohibits solicitation of professional employment when the lawyer knows or reasonably should know that the physical, emotional, or mental state of the prospective client is such that the person cannot exercise reasonable judgment. The opinion flags this as a real concern for automated intake tools. If your chatbot keeps cheerfully driving toward an engagement with a caller who is in acute distress, that is a 7.3(a) problem.

The practical answer is to build escalation triggers. Keywords, sentiment signals, or simply a default rule that flags long pauses and emotional language for human review before any engagement-related step occurs.

Two scenarios where this gets you in trouble

Scenario one: The accidental attorney-client relationship

A Bend family law solo deploys an AI receptionist on her firm's website. The vendor's product is configured to "engage the client" — including drafting and sending engagement letters when the conversation reaches certain milestones.

A man visits the site at 11 p.m. after his wife filed for divorce that afternoon. The chatbot, which sounds warm and competent, listens to his situation. It asks clarifying questions. It tells him the firm can help. It collects his payment information for the retainer and sends an engagement letter to his email before he goes to bed.

The attorney finds out Monday morning when the man calls asking when the temporary order will be filed.

Is there an attorney-client relationship? Quite possibly. Under Oregon law, the existence of an attorney-client relationship depends on the reasonable belief of the prospective client. In re Weidner, 310 Or 757 (1990). A signed engagement letter and a paid retainer are not nothing.

The fix is structural, not cosmetic. The chatbot must never represent that a relationship has formed. Engagement letters, fee quotes, and acceptance of representation must require explicit human approval. Disclaimers must be visible at the start of every chatbot session — not buried in a footer.

Scenario two: The disqualifying-information trap

A criminal defense solo in Bend uses an AI voice receptionist to handle after-hours calls. The system runs a thorough intake: name, charges, dates, witnesses, prior counsel, anything the caller volunteers.

At 9 p.m. on a Tuesday, the system takes a long call from a woman who is the sister of a complaining witness in a pending assault case. She is considering retaining the attorney to advise her about a potential civil claim related to the same incident. She gives the chatbot the full picture of what her family knows, what the prosecutor has told them, and what the witness will say at trial.

Two weeks later, the actual defendant in that assault case calls. The case is bigger. The fee is bigger. The attorney would very much like to take it.

She likely cannot. Under RPC 1.18(c), the firm received material information from a prospective client whose interests are materially adverse to the prospective defendant in the same matter. The imputation can be cured by written informed consent from both parties, but the practical reality is that the complaining witness's sister will not consent — she has no incentive to.

The attorney is conflicted out by a chatbot conversation she never heard.

The fix is to collect minimum viable intake on the front end — name, jurisdiction, opposing party, contact information — and route the rest of the conversation to a human conflict check before any substantive discussion occurs.

A vendor-evaluation checklist

If you are evaluating an AI intake tool, the questions below are tied directly to the obligations 2026-208 imposes on you. A vendor that cannot answer eight of these ten without checking with someone is not ready for your firm.

1. Does your AI identify itself as an AI program at the start of every conversation?
2. Can the system be configured so that engagement letters, fee quotes, and anything resembling acceptance of representation require explicit human approval before being sent?
3. How do you isolate one client's information from another's? Can prompts or conversations from my firm's deployment be used to train responses for any other customer?
4. What encryption do you use at rest and in transit, and can you produce documentation of your security controls — including SOC 2 Type II or comparable?
5. Will you disclose every sub-vendor in your stack — including any underlying language model provider, speech-to-text vendor, and storage vendor — and will you contractually bind them to confidentiality?
6. How will I be notified of every chatbot interaction, and how quickly?
7. Can I review and modify the chatbot's responses, and what is the audit trail?
8. What is your escalation protocol when a caller appears to be in crisis or emotionally impaired?
9. Will you sign a written agreement that includes confidentiality, breach notification, data return or destruction on termination, and your obligations as my nonlawyer assistant under RPC 5.3?
10. In the event of a data breach, what is your notification window, and what indemnification do you provide?

Copy these into an email to a sales rep. The answers — and the speed with which they come — tell you more than the demo.

A sample disclosure for your website or call greeting

Florida's Ethics Opinion 24-1 explicitly requires lawyers to inform prospective clients that they are communicating with an AI program. The New York City Bar's Formal Opinion 2024-5 reaches the same conclusion. Oregon's 2026-208 strongly implies the same outcome through RPC 7.1, even if it stops short of mandating specific language. Either way, you need a disclosure.

The following is a template you can adapt. It is not a one-size-fits-all clause, and it does not relieve you of the obligation to think about your specific situation:

This site uses an AI assistant to help us respond to inquiries twenty-four hours a day. The AI is not a lawyer and is not authorized to give legal advice, accept your case, or establish an attorney-client relationship. Any information you share will be reviewed by an attorney at [Firm Name] before next steps are taken. If your situation is urgent, please call us directly at [number], or call 911.

For a voice receptionist, the disclosure should be the first sentence the caller hears, after the firm name. Not after the system asks how it can help. First.

What the opinion does not cover

2026-208 is a state bar ethics opinion. It addresses your obligations under the Oregon Rules of Professional Conduct. It does not address — and should not be read to address — several adjacent issues that may still bite you:

• Federal court filing disclosure rules. The U.S. District Court for the District of Oregon may impose separate requirements about AI-assisted filings. Check standing orders before you file.
• TCPA compliance. The Telephone Consumer Protection Act regulates outbound AI calling separately, with penalties from $500 to $1,500 per call. An inbound AI receptionist is generally lower risk, but if your system makes outbound follow-up calls, you have a federal compliance regime to navigate on top of the bar opinion.
• Statutory consumer-marketing liability. Oregon and federal consumer protection statutes may apply to misrepresentations by your AI in ways the bar opinion does not address.
• Contract law. If your AI quotes a fee, makes a promise about timeline, or otherwise creates terms that turn out to be wrong, the resulting contract dispute is separate from any bar discipline issue.
• Blanket immunity. As Mark Fucile noted in his analysis for NWSidebar, the opinion focuses on professional conduct rules and does not address the full universe of risk management considerations.

How we think about this at WildRun AI

WildRun AI is a voice AI receptionist built for the kind of small firm 2026-208 is mostly concerned with. Two things we did differently because of how the opinion reads:

Our AI identifies itself as an AI at the start of every call, before the firm's greeting completes. We do not send engagement letters, quote fees, or do anything that could reasonably be construed as accepting representation. Every call produces a transcript that arrives in your inbox within minutes, and our supervision model is designed to work with trained staff handling first review — which footnote 10 of the opinion explicitly permits.

If you are an Oregon attorney evaluating AI intake — ours, a competitor's, or one you've already deployed — we offer a free twenty-minute review of your intake workflow against the requirements of 2026-208. You do not need to be a customer. We will tell you honestly whether what you are doing is defensible, and if it is not, what to fix. The review is genuinely useful regardless of which vendor you choose.

You can book it here.

Further reading

• Oregon Formal Opinion 2026-208 — full text (PDF)
• Oregon Formal Opinion 2025-205 — general AI use
• ABA Formal Opinion 512, July 29, 2024
• Florida Bar Ethics Opinion 24-1, January 19, 2024
• Mark Fucile, "New Oregon State Bar Opinion on Using AI Agents for Client Intake," NWSidebar (March 2026)

This article is for informational purposes only and does not constitute legal advice. WildRun AI is not a law firm and does not provide legal advice or representation. Attorneys evaluating any technology for use in their practice should consult their own ethics counsel and exercise their independent professional judgment.

Frequently asked questions