As artificial intelligence moves deeper into legal work, courts face a tempting shortcut: require everyone to disclose it.
A new guidance document from U.S. Magistrate Judge Maritza Dominguez Braswell offers a more useful distinction. The document does not ban AI. It also does not require every party to certify whether AI was used in a written submission. It is guidance, not a standing order.
That choice should not be mistaken for leniency.
The guidance says broad AI-use disclosures can create a false sense of security. A filing is not reliable merely because a lawyer checked a box about the tool used. The work still has to be accurate, supported, and responsibly reviewed. The guidance instead points back to the duties already attached to litigation: verification, candor, and accountability for the paper presented to the court.
That is the better question for a court, a law office, or a self-represented litigant to ask: not simply, “Was AI used?” but, “What was checked before this was filed?”
The document proposes a practical workflow. Start with your own reading and judgment. Use AI in the middle of the process to find gaps, identify inconsistencies, or surface counterarguments. Then verify the result outside the AI tool before finalizing anything that matters.
That sequence matters because asking a tool whether its own answer is correct is not verification. A real check means returning to a reliable legal or factual source and confirming that the cited case, quoted language, factual statement, and legal proposition actually hold up.
The guidance also recognizes that risk is not all the same. Formatting a lawyer's own notes does not carry the same stakes as preparing a filing, reviewing discovery, or handling sensitive case materials. As the consequences rise, the level of review should rise with them.
This is a useful model for policy design because it avoids two bad answers.
The first bad answer is a blanket prohibition. Courts and lawyers already encounter AI inside common tools, and a formal ban may be hard to define, harder to administer, and poorly fitted to different kinds of work. The second bad answer is a blanket certification requirement that treats disclosure as if it proves accuracy.
Neither answer removes the human duty.
A litigation system does not need to pretend that every use of AI is the same. It needs a clear rule for responsibility. The person who signs, files, or advocates a paper remains responsible for what it says. That is why the most durable safeguard is not a label attached at the end of a document. It is the work done before the document reaches the court.
For Tennessee lawyers and court leaders, the practical lesson is simple. Build the review path into the workflow: identify the task, assess the risk, preserve the source material, require an independent check, and make sure the responsible person can explain the final product in plain language.
The question for this week
A checkbox can disclose a process. It cannot perform the verification.