The duty to understand your AI is becoming mandatory. Here is the checklist.

What actually happened in June 2026

Three things landed in the same stretch of June 2026, and together they form a pattern worth your attention.

California revised its proposed AI ethics rules after a round of public feedback, as reported by Bloomberg Law. The Daily Journal, covering the California proposals, said they "put attorneys on notice." Around the same time, Rhode Island issued new AI rules for lawyers, reported by JD Journal on June 18, 2026, and Ohio published an AI ethics guide, noted by Husch Blackwell in June 2026.

No single one of these is a thunderclap. Read together, they are a direction. State after state is writing down what it expects when a lawyer uses generative AI, and the common thread is that you are responsible for the output, full stop, in the same way you are responsible for the work of a junior associate.

Why this is the competence duty, not a new rule

The instinct is to treat each new guidance document as a separate compliance task. That is the wrong read. Most of this guidance is not inventing a new obligation. It is applying the existing duty of competent representation to a new tool.

The duty of technological competence has been on the books in many jurisdictions for years, often as a comment that lawyers should keep abreast of relevant technology. For a long time it sat there as something aspirational, the kind of line you nod at and move past. What the 2026 wave does is make it operational. It signals that regulators now expect you to actually understand the AI tools you use well enough to supervise them.

That is the shift. Aspirational becomes mandatory. The question a bar might ask is no longer "did you mean well" but "what did you do to check the work."

It also explains why the guidance is arriving from so many directions at once. No state is waiting for a model rule before saying what it expects, because the underlying duty is already theirs to enforce. California revising after public feedback, Rhode Island writing its own rules, Ohio publishing a guide: these are not coordinated, and they do not need to be. Each is reaching the same place from the same starting point. When you see that pattern, the safe assumption is that your jurisdiction either has its own version already or will soon. Waiting for perfect clarity is not a strategy, because the duty applies now even where the guidance is still being written.

What competent supervision of AI looks like, written down

Here is the part that matters for your Tuesday. Competent supervision is a set of concrete habits. A small firm can adopt this version this quarter.

• Know what the tool is and is not. Before AI touches client work, you should be able to say in one sentence what the tool does, where it tends to fail, and that it can produce confident, wrong output. You do not need to be an engineer. You need an accurate working model.
• Supervise every output as you would a junior's draft. Read it. Check anything that would matter if it were wrong. Verify every citation, name, and figure against the source, not against the model's own confidence.
• Keep a record of the check. A short note in the file, "AI used for first draft, citations verified against originals on this date," is the difference between a habit and a habit you can prove.
• Set a firm policy, even a one page one. What tools are approved, what may never go into them, who is accountable for review. A written policy is itself evidence of supervision.
• Re-check when the tool changes. Models update. A workflow that was safe last quarter deserves a fresh look when the underlying tool shifts under you.

None of this is exotic. It is the supervision you already apply to delegated work, pointed at a new kind of delegate.

The reason to write it down rather than just do it is evidentiary. A habit that lives only in your head is invisible to a regulator and to a worried client. A habit that leaves a trail, a one line policy, a short note in each file, an approved-tools list, is a habit you can demonstrate. The cost of that documentation is a few minutes. The value of it shows up exactly once, on the day someone asks what you did, and on that day it is the difference between a calm answer and a scramble. Build the trail when there is no pressure, so it is already there when there is.

A practical sequencing tip for a small firm: do not try to roll all five out at once. Start with the review-and-record habit on the next piece of client work AI touches, because that is the one that protects an actual client today. Add the written policy this month. Layer in the periodic re-check as a calendar reminder. The goal is a practice you actually run, not a binder you write once and never open.

The trap of the confident draft

There is one honest caveat that sits above all the checklists. A more capable model produces more convincing output, and convincing output is harder to supervise, not easier. When a draft reads beautifully and cites cases by name, the pull is to trust it. That pull is the risk.

The discipline that satisfies the competence duty is also the one that protects your client: assume the polished draft contains an error until you have checked it yourself. Treat fluency as a reason for more scrutiny, not less. This is the single habit that the whole wave of guidance is, in effect, asking you to write down and follow.

For the question of what you may and may not feed these tools in the first place, that is a confidentiality question with its own answer, and it deserves separate treatment.

Where billing fits, and where it does not

A quick boundary, because the two get tangled. How you charge for AI-assisted work, whether you can bill for time the tool saved, how you describe it on an invoice, is a real and separate ethics question. We cover that in Billing for AI-assisted legal work.

This briefing is about the duty before the bill: understanding and supervising the tool so the work itself is sound. Get the competence and supervision right first. The billing question is easier to answer cleanly once the underlying work is defensible. And if you are the senior person setting the standard for a team, the senior lawyer AI operating model is the companion piece on making this stick across a firm.

The skill under the rules

Every state that publishes guidance is, in its own words, describing the same thing: a lawyer who understands the tool and can show the work. That is not a document you download. It is a method you build and run.

The advantage was never access to the model. Everyone has that. What separates the lawyers who stay ahead of this is a repeatable practice for using AI on real client work without losing control of it. That is what The Leveraged Attorney teaches from the ground up, and the two minute course quiz will point you to the right starting place for your practice.