Most LNCs Are Using AI to Save Time. The New Federal Rules Are About to Make That a Liability.

Most consultants using AI right now are asking the wrong question. They're asking "does this save me time?" The question that actually matters is: "can I defend this under cross-examination?"

Those are not the same question, and the gap between them is about to get a lot more expensive.

What's actually changing

In June 2025, the Advisory Committee on Evidence Rules approved a new Federal Rule of Evidence — Rule 707 — aimed specifically at machine-generated evidence. The rule is short, and it says exactly what it means: when machine-generated output would be subject to Rule 702 if a human expert had said it instead, the court can only admit it if it independently satisfies Rule 702's reliability requirements. It went through public comment ending in February 2026 and is still moving through the rulemaking process — it isn't binding law yet, but it tells you exactly where federal evidentiary standards are headed, and several courts are already reasoning this way ahead of formal adoption.

There's a companion amendment to Rule 901 on authentication. If AI-generated evidence gets challenged, the burden shifts to whoever's offering it to prove it's more likely than not authentic before it comes in at all.

Translation: the era of "the AI said so" quietly riding into a case unexamined is ending. If it hasn't ended in your jurisdiction yet, it's coming.

A case worth knowing cold

In In re Celsius Network, an expert witness used generative AI to produce a 172-page valuation report in 72 hours — a report he admitted would have taken a human over 1,000 hours to write. The court excluded it. Not because AI was involved. Because the report contained duplicated paragraphs, factual errors, and almost no citations back to the underlying data. The expert hadn't actually reviewed the source material the report claimed to analyze. His live testimony was still allowed, because he could speak to his own reasoning — the report couldn't speak for itself, and neither could the machine that wrote it.

That's the whole lesson, and it applies to you directly, not just to attorneys. Speed isn't the problem. An unverifiable methodology is the problem.

Why this is an LNC issue, not just an attorney issue

Attorneys aren't the ones running the AI tool during record review. You are. If you use AI to summarize a chart, draft a first-pass chronology, or flag potential gaps — and any of that language ends up in your findings without you independently verifying it against the source record — you are the one who built an unreliable methodology into your own work product. Under your own name. On a report you may eventually have to defend under oath.

A few standards I'd hold every LNC to, starting now:

Never submit AI output as your own clinical opinion without independently verifying it against the source record. If you can't point to the specific page and entry that supports a conclusion, it doesn't go in the report — whether a human or a machine drafted the sentence.

Be able to explain your reasoning without referencing what the AI told you. If the honest answer to "how did you reach that conclusion" is "the tool said so," you have already lost that exchange before the follow-up question.

Treat AI output the way you'd treat a rough first draft from a junior reviewer you don't fully trust yet. Useful for speed. Not citable. Not a substitute for your own read of the record.

Disclose your process, don't hide it. If a tool materially assisted your review, decide now — before you're asked under oath — how you'll describe that honestly and defensibly.

What this looks like in my own practice

I use AI daily. I've built a database that lets me query and cross-reference records fast — flag patterns, pull entries by provider or timeframe, cut down the hours it takes to find what matters across a few thousand pages. I use it to identify what metadata I actually need to pull for a given case and where to find it — access logs, modification timestamps, audit trail fields — before I go digging manually. I use it to generate a first-pass timeline. I use it to run statement analysis on documentation and interview text, flagging language worth a closer look.

None of that touches my findings until I've verified it myself, against the source record, line by line. Every timeline I put my name behind has been manually confirmed — not spot-checked, confirmed — before it's mine to stand behind. AI runs as my second or third check on a conclusion I've already reached independently. It never runs as the source of the conclusion itself.

That's the actual distinction the new rules are drawing, and it's the distinction that was always going to separate a real LNC from a liability — not whether you use AI, but where it sits in your process. Front of the line, making the call? That's a Celsius Network moment waiting to happen with your name on the report. Back of the line, catching what you might have missed after you've already done the work independently? That's just a well-built process with better tools in it.

Why this is actually good news, if you're doing this right

Every one of these new rules is drawing a sharper line between an accountable professional and an unaccountable process. That line is exactly where a real LNC's value has always lived. A machine can flag that a vital sign trended downward. It cannot tell you whether that trend should have triggered an intervention, whether the documentation pattern around it looks like a busy shift or a missed one, or whether the gap actually matters clinically. It cannot be deposed. It cannot be cross-examined. It cannot be held accountable — professionally, ethically, or legally — for being wrong.

You can. That's not a limitation. That's the entire reason you're the one attorneys hire instead of running the record through software themselves.

AI is a tool in this work, the same way a chronology template or a records database is a tool. The moment you start treating its output as an opinion instead of a starting point, you've stopped being the expert and started being a pass-through for something that can't stand behind its own conclusions in a courtroom. The new rules aren't a threat to good LNC work. They're a threat to LNC work that was already resting on a foundation that couldn't survive real scrutiny.

Use the tool. Just make sure you're still the one doing the thinking.

One more standard, and it's the one underneath everything else in this piece

I don't reserve this scrutiny for AI. I apply it to every input that touches my work — the subject matter experts I consult, the texts and literature I reference, the methodologies I rely on to reach a conclusion. Before any of it makes it into a report with my name on it, I ask the same question a Daubert hearing would ask: has this been tested, is it generally accepted in the relevant field, is the error rate known, would it survive someone actively trying to take it apart?

Florida adopted Daubert as its evidentiary standard in 2019, replacing the older Frye "general acceptance" test — and whether I'm working a Florida case or one in a state that still applies Frye, I hold my own inputs to that standard by default. Not because a court might someday ask me to. Because it's the only standard that actually protects the work before it ever gets that far.

AI is simply the newest thing that has to clear that bar. It was never going to get a pass that a textbook, a colleague's opinion, or my own first instinct doesn't also have to clear.

Shane Huey, MS, MBA, RN, LNC — legal and forensic nurse consultant with an 11-year background building and maintaining EHR systems, and an MS in Information Security focused on HIPAA-regulated systems and digital forensics.