Back in May 2025, the U.S. Judicial Advisory Committee on Evidence Rules proposed a new draft Rule to the Federal Rules of Evidence. The Rule purportedly would regulate how AI-generated evidence is to be introduced at federal trials. It opened the Rule up for comment and on January 15, 2026, a hearing was held. Suffice it to say there was not much support — for once, plaintiffs class action lawyers and in-house legal counsel both objected. And for good reason: the proposed rule addresses a problem that’s not a problem while ignoring the very real challenges facing the judiciary.
The Proposed Rule
The draft Rule in question, Rule 707, states as follows:
When machine-generated evidence is offered without an expert witness and would be subject to Rule 702 if testified to by a witness, the court may admit the evidence only if it satisfies the requirements of Rule 702(a)-(d). This rule does not apply to the output of simple scientific instruments.
Rule 702 deals with the standards by which expert testimony and evidence is to be determined to be admissible. Generally speaking, the evidence must be helpful to the trier of fact, be based on reliable principles, and constitute a reliable application of those principles. It grew out of a Supreme Court ruling in 1993 in Daubert v. Merrell which set standards for how courts are to assess expert testimony admissibility. The result is frequent hearings challenging the admissibility of expert testimony in court.
According to the Committee Notes on the proposed Rule, the idea is that the same standards for admissibility should apply to “machine-generated evidence” offered by a layperson as to that offered by an expert. The Committee reasoned that a lay witness might apply a “program” but know little or nothing about its reliability. The Committee seemed to be thinking about a technician who would enter a question and then print out the answer and then testify about what they found.
It appears that this Rule is one of the first efforts to deal with presumably AI-generated evidence.
Rule 707 Ignores the Real Issues
What is alarming is that the Rules Committee is focusing on something that is not that big of a problem particularly for experienced lawyers and judges when there are plenty of bigger issues facing the judiciary. Like the proliferation of deepfakes about which I have written several times. Like the threats of AI to judicial decision making and how judges should be using it, also a subject I and others have addressed. Like the cost of litigation that freezes so many people out of courtrooms. Like bias in the AI tools that are being used every day. Like the lack of technological training and understanding many judges desperately need to conduct fair trials. Like access to justice. Like threats to the rule of law itself.
If you want a good understanding to the threats to the judiciary and what can be done about, I suggest you read the comments of Dr. Maura Grossman, Hon. Herbert Dixon, Hon. Allison Goddard, Hon. Xavier Rodriguez, Hon. Scott Schlegel, and Hon. Samuel Thumma in the section entitled AI and the Courts in the recent ABA Report on the Impact of AI on the Practice of Law. I know most of these judges and experts: they are truly the cutting-edge thinkers on the state of the judiciary today.
The proposed Rule focuses on none of these things. Nor have any other proposed Rules.
Rule 707 Ignores Courtroom Practicalities
Beyond ignoring serious problems, Rule 707 creates new ones and ignores practicalities.
First and foremost, what the hell is machine-generated evidence? A calculator is a machine and can generate numbers used as evidence. So can a spreadsheet on a computer. So can a lot of “machines” we use in daily life. The Committee tries to save itself by adding the last sentence which provides that the Rule won’t apply to simple scientific instruments. But again, what does that mean? What is considered simple? For that matter, what is an instrument? Not to mention the fact that what is considered a simple instrument today once upon a time was far from it.
Certainly, the rule is aimed at GenAI. But the Committee ought to just say that and provide a definition. Otherwise, it just muddies the water.
Another faulty presumption: experts who testify have more knowledge about the reality of machine-generated evidence than anyone else. The truth is some experts may have knowledge of this kind of evidence (whatever it is) but many don’t. An accident reconstruction expert may know about how to reconstruct an accident but not how ChatGPT created the drawing and image they want to use.
And courtroom practicalities suggest that we don’t really need this kind of ambiguity. If a lay witness wants to use evidence generated by ChatGPT or anything else for that matter, most lawyers request a voir dire examination in front of just the judge and, if necessary, would then argue the witness is not qualified to talk about or rely on something they know little about. Why get into whether the evidence is machine-generated?
A lay witness is not an expert and can’t render opinions on technical issues for which they aren’t qualified. We don’t need a separate rule.
And our trials (when we have them — which isn’t often, at least on the civil side, as I have discussed) are already too costly and time-consuming. Under this new Rule, we would now have to have a separate Rule 702 or Daubert type inquiry for lay testimony when admissibility is easy to resolve. I’ve been in those Daubert fights. They are long and often pointless.
Plus, it opens the door to all sorts of attacks on lay testimony based upon the claim the witness used some sort of evidence that was generated by a machine. More attacks, more time spent on things that are obvious.
Experienced judges can deal with admissibility of lay-witness testimony. They don’t need more unclear rules for lawyers to waste their time arguing about.
And if we don’t have enough experienced judges, then let’s try to solve the real problem.
Many similar sentiments were expressed at the January 15th hearing.
Both Sides Question the Proposed Rule
According to a Reuters article reporting on the hearing, those attending generally made many of the same points. The interesting thing is that both sides of the aisle seemed to question whether the rule was a good idea right now, particularly when technology is changing so rapidly. Plaintiffs’ lawyers worried that it could increase cost especially in complex litigation. The in-house legal representatives worried that the rule would create uncertainty and was overly broad since it could sweep in routine uses of AI tools.
Both sides worried that it would result in more expert testimony. This is a particularly valid point. Experts are great but, in my experience, there is usually an expert somewhere, someplace to opine on virtually anything, at the right price. The result is the process is gummed up and juries are left confused and uncertain when presented with diametrically opposed opinions but two seemingly qualified experts. We don’t need more of that.
Taking the Easy Way Never Works
These are lots of hard topics we need to address as set out above. Many may be controversial. But they are real. Taking the easy way out and offering Rules like 707 that aren’t needed seems like an effort to appear to be doing something while avoiding things we don’t want to talk about.
That’s not a solution. Lawyers and judges should unite and talk about real problems and solutions that challenge us both if we are interested in the rule of law. Because the real problems confronting us in the new age of AI, if not handled correctly and soon, may threaten the rule of law bedrock on which we all depend.
Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes TechLaw Crossroads, a blog devoted to the examination of the tension between technology, the law, and the practice of law.
The post Proposed Evidentiary Rule 707: Addressing A Nonexistent Problem Instead Of Real Ones appeared first on Above the Law.