There is a lot of confident writing about new evidence rules for AI. Some of it gets the one thing that matters wrong, which is whether the rule exists yet. Headlines have described a proposed rule as approved when what had been approved was publishing it for public comment.

For an investigator, that difference is not academic. Cite a rule that is not in force, in a report or on the stand, and you have handed the other side the easiest impeachment of the day.

So here is where things stand as of September 2026, from an investigator's side of the table. I am not a lawyer. Everything below is cited so you can read the source yourself, and you should.

What is in force: Rule 702, as amended in 2023

The rule that governs expert testimony in federal court was amended effective December 1, 2023. It now reads that a qualified expert may testify "if the proponent demonstrates to the court that it is more likely than not that":

(a) the expert's scientific, technical, or other specialized knowledge will help the trier of fact to understand the evidence or to determine a fact in issue; (b) the testimony is based on sufficient facts or data; (c) the testimony is the product of reliable principles and methods; and (d) the expert's opinion reflects a reliable application of the principles and methods to the facts of the case.

Fed. R. Evid. 702.

Two things changed. The "more likely than not" language made explicit that the proponent carries the burden on each element by a preponderance. And subdivision (d) changed from asking whether the expert "has reliably applied" his methods to whether his "opinion reflects a reliable application" of them.

The committee gave two reasons. Some courts had been treating questions about the sufficiency of an expert's basis and the application of his method as matters of weight for the jury, rather than admissibility questions for the judge. And the change to (d) was aimed at overstatement: experts, forensic experts in particular, drawing conclusions stronger than their method could support.

The framework it builds on goes back to Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993), which made the trial judge the gatekeeper for the reliability of expert testimony.

Why an investigator should care. Subdivisions (b) and (d) are exactly where AI-assisted work gets tested, when an investigator is offering expert testimony. Was the opinion based on sufficient facts or data, or on a tool's summary of them? Does the opinion reflect a reliable application of the method to these facts, or did a tool's output stand in for the application? The overstatement concern matters here too. AI drafting tools have a habit of strengthening the language of a finding, and an opinion that says more than the work supports is precisely what the 2023 amendment was written to catch.

An investigator testifying as a fact witness is governed by different rules. But he cannot avoid Rule 702 simply by calling himself a lay witness. Lay opinion is limited to testimony "not based on scientific, technical, or other specialized knowledge within the scope of Rule 702." Fed. R. Evid. 701(c). If his conclusion depends on specialized method, the court can hold it to the expert standard.

Arizona

Arizona adopted the federal standard by order of the Arizona Supreme Court, effective January 1, 2012. Ariz. Sup. Ct. Order R-10-0035 (Sept. 7, 2011). That replaced the state's earlier approach under Logerquist v. McVey, 196 Ariz. 470, 1 P.3d 113 (2000).

Arizona then conformed its rule to the 2023 federal amendment, effective January 1, 2024, adopting both the preponderance language and the revised subdivision (d). Ariz. Sup. Ct. R-23-0004. In Arizona state court, the standard above is the standard.

Other states have not all moved together. If you testify outside Arizona or federal court, check that state's current rule rather than assuming it matches the federal text.

What is proposed and not in force: Rule 707

This is the one that gets misreported.

In 2025 the federal rules committees published a proposed new Rule 707 for public comment. As published, it read:

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.

The idea is straightforward. If a person offering the same conclusion would have to satisfy Rule 702, a machine offering it should not get around that just because no human is on the stand.

The public comment period ran from August 15, 2025 to February 16, 2026. The committee described the response as "mixed, but helpful." Only 3 comments expressed unqualified support, 27 supported the rule subject to revisions, and 27 opposed it. The most common objection was breadth: "Most commentators thought that the term 'machine-generated' was overbroad," reaching routine machine output such as geolocation data. Others noted that it does nothing about evidence whose authenticity is in dispute, since it applies to evidence the proponent acknowledges is machine-generated. Others raised the cost of the technical experts it would require. The Department of Justice cast the only vote against publishing it, taking the position that Rule 702 already covers machine-generated evidence.

In May 2026 the Advisory Committee on Evidence Rules declined to send the rule forward: "The Committee does not recommend action on the proposed Rule 707 at this time." It had already redrafted the rule substantially in response to the comments, so the text quoted above is no longer the committee's working version. Rather than publish the revision for a second round of comment, it chose to have it vetted by technology experts at its fall 2026 meeting. Advisory Comm. on Evidence Rules, Report to the Comm. on Rules of Practice and Procedure (May 17, 2026).

Rule 707 is not a rule. Do not cite it as one. It may come back narrower, broader, or not at all, and the version that comes back will not be the one most articles quote.

Deepfakes: not a rule either

The committee has also considered how to handle evidence alleged to be a deepfake, under a working draft of Rule 901(c). Its conclusion in May 2026 was direct: "The Committee is of the view that, at least for now, an amendment to Rule 901 to address deepfakes is not warranted." Id. The subject stays on its agenda for the fall.

So disputed media still runs through the existing authentication rule, which requires the proponent to "produce evidence sufficient to support a finding that the item is what the proponent claims it is." Fed. R. Evid. 901(a). One of the rule's own illustrations is worth an investigator's attention: "Evidence describing a process or system and showing that it produces an accurate result." Fed. R. Evid. 901(b)(9).

That illustration is a fair summary of what any AI-assisted method has to be able to survive. Describe the process. Show that it produces an accurate result.

What this means for an investigator right now

The existing rules already reach you. If your work supports an expert opinion, or you are testifying as one, Rule 702's requirements apply today. If your opinion rests on specialized method, Rule 701(c) can bring you under them even as a fact witness. You do not get to wait for Rule 707.

Machine output is not a finding. Whether or not Rule 707 is ever adopted, the logic behind it holds: a conclusion does not become more reliable because a machine produced it. Treat every AI output as a lead that you verified against a primary source, and be able to say which source.

Document the method, not just the result. Rule 901(b)(9) and Rule 702(d) are both, in the end, questions about process. An investigator who can describe his process and show the checks that produced his findings is answering the question the rules are asking.

Know where your competence stops. You can document how submitted media was preserved, hashed and handled. Whether it is authentic is a forensic examiner's opinion. Saying so on the record is not a weakness. It is the kind of limit a court expects a reliable witness to know.

Watch the fall meeting. The committee's next steps on Rule 707 and on deepfake authentication are expected there. When it moves, I will write it up here, with the citation.

The verification standard referred to here is set out in Chapter 3 of The AI-Powered Private Investigator, and testimony preparation in Chapter 13. I review AI-assisted investigative work for counsel and testify about how it was done. Expert witness and consulting.

Jon Jacobson, MBA, is a licensed Arizona private investigator (license #1799200) and principal of Old Pueblo Investigations in Tucson. He is not an attorney, and this is not legal advice. Rules of evidence differ between federal and state courts and change over time. Status stated as of September 2026. Confirm the current rule in your jurisdiction before relying on it.

Authorities

  • Fed. R. Evid. 702 (as amended Dec. 1, 2023). Text
  • Fed. R. Evid. 702 advisory committee's note to 2023 amendment (preponderance clarification; revision to subdivision (d) addressing overstatement). Text
  • Fed. R. Evid. 701(c) (lay opinion limited to testimony "not based on scientific, technical, or other specialized knowledge within the scope of Rule 702"). Text
  • Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993). Opinion
  • Ariz. R. Evid. 702; Ariz. Sup. Ct. Order R-10-0035 (Sept. 7, 2011) (effective Jan. 1, 2012) (adopting Fed. R. Evid. 702 as restyled). Order
  • Ariz. Sup. Ct. R-23-0004 (amending Ariz. R. Evid. 702 to conform to the 2023 federal amendment; effective Jan. 1, 2024). Arizona Supreme Court 2023 Rules Summary
  • Logerquist v. McVey, 196 Ariz. 470, 1 P.3d 113 (2000). Opinion
  • Proposed Fed. R. Evid. 707, in Comm. on Rules of Practice and Procedure, Preliminary Draft of Proposed Amendments to the Federal Rules (Aug. 15, 2025) (comment period closed Feb. 16, 2026). Not adopted. Publication
  • Advisory Comm. on Evidence Rules, Report to the Comm. on Rules of Practice and Procedure (May 17, 2026) (not recommending action on proposed Rule 707; characterizing public comment; concluding that an amendment to Rule 901 to address deepfakes "is not warranted" at least for now). Report
  • Fed. R. Evid. 901(a), (b)(9). Text