by Frank Hudson


You’ve probably already used AI for legal research. Maybe you asked Harvey AI, Anthropic’s Claude Legal, or Google Gemini to summarize a statute, or had it draft a quick memo on a procedural issue. If the output looked good, you might have copied it straight into your work product.

Bad idea. An Oregon lawyer just found out the hard way that AI-generated citations can cost you $2,000 in sanctions and seriously damage your credibility with the court. The December 2025 decision in Ringo v. Colquhoun Design Studio, LLC is your wake-up call.

When AI Goes Wrong: A Cautionary Tale from the Oregon Court of Appeals

In December 2025, the Oregon Court of Appeals issued a stark warning in Ringo v. Colquhoun Design Studio, LLC, 345 Or App 301 (2025). In Ringo, an attorney filed a brief “littered with fabricated cases, a fabricated quotation, and fabricated substantive law.” The court struck the brief and imposed $2,000 in sanctions: $500 for each false citation and $1,000 for each false quotation or statement of law.

The Ringo decision highlights what can go wrong when lawyers rely on AI without verification. While the attorney never explicitly acknowledged using artificial intelligence, the court recognized the reality: the fabricated law “likely resulted from the use of artificial intelligence.”

Chief Judge Lagesen’s opinion rejected the industry-standard term “hallucination” as fundamentally mischaracterizing AI errors. Citing Merriam-Webster’s Collegiate Dictionary, the court observed that hallucinations are typically caused by nervous system disorders or the consumption of drugs like LSD. AI systems, by contrast, do not experience such conditions. Instead, the court noted, AI generates false citations “in accordance with its design,” a distinction that reframes the error from symptom to systematic output.

And yes, the court really did spend a paragraph contrasting AI outputs with the effects of LSD or a nervous system disorder. But the distinction is vital: there isn’t a bug or a drug in the system. It is exactly how these tools work. This judicial reframing suggests we need more precise language.

Instead of “hallucination,” the more technically correct term is “confabulation,” which occurs when a large language model fills gaps in its knowledge by generating plausible-sounding but false information. While the court used the terms ‘fabricated’ and ‘nonexistent,’ the mechanism it described is exactly that: confabulation. Stated differently: people hallucinate, but large language models confabulate. Consequently, the court in the Ringo decision imposed sanctions on respondents’ counsel and struck his brief because the brief contained AI-confabulated legal authorities.

The court identified three reasons why submitting fabricated legal authority is “exceptionally grave:”

First, it breaches professional duties. Oregon RPCs 3.3 (Candor) and 4.1 (Truthfulness) prohibit false statements of law to tribunals and third parties.

Second, it strains judicial resources. Every hour spent verifying false citations is an hour diverted from deciding cases briefed under law that actually exists.

Third, it jeopardizes the rule of law. As the court noted, “judicial precedent is the backbone of the rule of law.” The injection of false precedent shakes that foundation.

The Ringo court also imposed an unprecedented remedy for future filings: any brief from that attorney must contain a certification that counsel drafted the brief without using generative AI to produce a draft, has read each cited case, and has verified that every source exists.

This isn’t just a cautionary tale: it’s a preview of what Oregon lawyers can expect if they fail to use AI responsibly. So what are you supposed to do about it?

What Oregon Law Requires: The New Framework

The Oregon State Bar anticipated these problems. In February 2025, just months before Ringo, the Board of Governors approved Formal Opinion No. 2025-205, establishing clear guidelines for AI use. The opinion reinforces that Oregon lawyers may use AI tools, but only if they take reasonable steps to remain competent in the technology and comply with existing Rules of Professional Conduct.

In practice, this means:

Competence (RPC 1.1): You must understand your AI tool’s capabilities and limitations before relying on its output. This is an ongoing obligation as these tools rapidly evolve.

Confidentiality (RPC 1.6): You must vet AI vendors’ security practices and ensure they don’t use client data for training without consent.

Supervision (RPC 5.1-5.3): You are ultimately responsible for all AI-generated work product, just as you would be for work by a junior associate.

Candor (RPC 3.3): You must independently verify every citation. Research cited in Formal Opinion 2025-205 suggests general-purpose AI models “hallucinate” answers to legal questions regarding a court’s core rulings around 75% of the time (see note 14 of Formal Opinion 2025-205).

The reality is that AI doesn’t reduce your ethical obligations. If anything, it expands them.

Three Essential Practices for Ethical AI Use

Making the shift from knowing the rules to following them requires a practical framework. Here are three essential practices every Oregon lawyer should implement immediately.

1. Treat AI Output as a First Draft, Never a Final Product

AI is best understood as a highly efficient but unreliable research assistant. It can accelerate your work, but it cannot replace your legal judgment.

What verification means in practice:

  • Pull every cited case from an official source (Decisis, Westlaw, Lexis, or official court reporters)
  • Read the full opinion, not just the AI-generated summary or snippet
  • Confirm that quotations are accurate and not taken out of context
  • Verify that the legal principles stated actually appear in the cited authority

Formal Opinion No. 2025-205 is unambiguous: “[T]o avoid violation of Oregon RPCs 3.3 and 4.1, lawyers must review for accuracy any [generative AI] output discussing case-specific facts or providing a case citation, quotation, or conclusion.”

If you’ve previously submitted AI-generated content without thorough verification, consider adopting the certification practice the Ringo court suggested: personally certifying that you’ve read each cited case and verified its existence.

2. Understand Your AI Vendor’s Data Practices Before Inputting Client Information

Not all AI tools handle data the same way. Understanding the difference between “open” and “closed” large language models is critical to protecting client confidentiality under RPC 1.6:

A legal-specific large language model is an artificial intelligence system designed specifically for the legal domain. It is trained on extensive legal materials including case law, statutes, legal commentary, and practice documents. The system learns to recognize and understand legal language, concepts, and reasoning patterns, enabling it to assist with legal tasks such as document drafting, legal research, contract analysis, and memoranda preparation.

Open models continuously learn from data inputs, potentially incorporating and later disclosing sensitive client information you provide. If you use an open model with client data, you must obtain informed consent after explaining the material risks—including potential disclosure and loss of confidentiality or intellectual property rights.

Closed models process data in isolation within a secure environment, preventing external utilization. While closed models offer better confidentiality protection, you still need to carefully review vendor contracts.

Key contract provisions to examine:

  • How data is handled, encrypted, stored, and destroyed
  • Whether the vendor uses client inputs to train the model
  • What security protocols protect against unauthorized disclosure
  • Whether data is shared with external systems or entities

Even with a closed model, you may need to anonymize or redact sensitive information. And remember: simply removing a client’s name from a fact pattern may not be enough if the circumstances are unique enough to identify them.

3. Create Written Policies and Train Everyone Who Uses AI

If you have managerial or supervisory responsibilities, RPC 5.1 and 5.3 require you to ensure that everyone in your firm uses AI in compliance with the Rules of Professional Conduct.

Managerial lawyers must:

  • Establish clear written policies on permissible AI use
  • Specify which tools have been vetted and approved
  • Document the vetting process for each approved tool

Supervisory lawyers must:

  • Provide training on the specific AI tools being used (not just “AI in general”)
  • Ensure subordinates understand the difference between open and closed models
  • Train staff on best practices for secure data handling, privacy, and confidentiality
  • Make clear that subordinate lawyers must not follow directions to use AI in ways that violate their professional obligations

Think of these policies as you would your firm’s conflict-check procedures or trust account protocols—essential infrastructure that protects both you and your clients.

Document your compliance efforts. If an ethics question arises later, contemporaneous written policies and training records demonstrate you took reasonable precautions. Finally, what to do if despite your best efforts, AI-confabulated legal authorities wind up in a pleading, motion, or brief that you’ve filed with the court? To state the obvious, you should come clean with the court as soon as possible.

Looking Forward: AI Competency as the New Standard of Care

Legal AI has become a permanent fixture in legal practice. Clients are building efficiency expectations into their contracts, and courts in jurisdictions like New York and Illinois are establishing formal guidelines for AI use. The Ringo decision and Formal Opinion No. 2025-205 implicitly acknowledge these new realities while establishing clear boundaries.

Within a few years, proficiency with these tools will likely be as essential as traditional competencies like conflict checking. Lawyers who develop responsible AI practices will gain significant efficiency advantages, and those who don’t risk falling behind as client expectations evolve.

These guidelines aren’t designed to discourage AI adoption, they’re meant to ensure it’s done responsibly. Use these tools to work smarter, but don’t allow AI to replace your professional judgment. The AI doesn’t have a bar license. You do. Because when something goes wrong, it’s your name on the brief, your credibility on the line, and your client who pays the price.

AI can make you more efficient, but becoming a better lawyer? That part’s still on you.

____

By Frank Hudson, Attorney at Law | [email protected]

This article was drafted with the assistance of Claude AI and Google NotebookLM. Images by NotebookLM.

This article is for educational purposes only and does not constitute legal advice. Reading this post does not create an attorney-client relationship with Frank Hudson.