# An official website of the United States Government. 11111s > Briefs, arguments, decisions, and more. URL: https://www.frixlaw.com/law-library/documents/federal_court_rules%3Aus-national%3Abc1a412bc6d9320d ## Record - **Collection:** Court rules - **Document type:** Court rule ## Text An official website of the United States Government. 11111s 100 times worse for the lowest-accuracy demographic group than for the highest-accuracy demographic group).”). 17 A Pew Research survey found that half of all respondents are more concerned than excited about the use of AI in their daily lives, up from 38% just three years ago. The concerns fall into several categories, including anger over the lack of choice to use AI. See Brian Kennedy et al., How American View AI and Its Impact on Society, PEW RSCH. CTR. (Sept. 17, 2025); Shira Ovide, Americans Have Become More Pessimistic About AI: Why?, WASH. POST (Oct. 7, 2025). 18 AI experts have a more favorable opinion of AI than the general public, but they still don’t exactly understand how it works. Pew Research Center found that “[f]ully 56% of AI experts surveyed say AI will have a very or somewhat positive impact on the United States over the next 20 years. This compares with 17% among the general public.” Colleen McClain et al., How the U.S. Public and AI Experts View Artificial Intelligence, PEW RSCH. CTR. (Apr. 3, 2025); Mark Bailey, How Can We Trust AI, If We Don’t Know How It Works, SCI. AM. (Oct. 3, 2023) (“AI systems have a significant limitation: Many of their inner workings are impenetrable, making them fundamentally unexplainable and unpredictable.”). Page 6 of 10 witnesses and many top-notch experts who are beginners when it comes to machine learning and AI. 19 Does this mean that a rule isn’t necessary? Perhaps not. It may be that a specific rule is needed to address machine learning. It seems short-sighted to propose a rule that excludes technically savvy experts, sweeps in lay witnesses who may not have the knowledge to discuss certain reliability issues related to machine learning and algorithms, while also requiring courts to engage in gatekeeping. The risk of related satellite litigation over very basic tools and instruments that have been reliably admitted as evidence outside of the scope of a Rule 702 analysis seems high and could needlessly bog down trial courts. Finally, the prongs of 702(a)-(d) are not easily applied to AI or machine learning tools. It would be better to draft a rule that doesn’t require the application of another rule. On this point, LCJ and AAJ are in rare agreement. 20 Not only are the prongs of 702(a)-(d) hard to apply to machine outputs, but there is also a risk of creating diametric court decisions and circuit splits. IV. How to Improve the Proposed Rule AAJ recommends that the Evidence Committee go back to the drawing board and answer the basic questions of what needs to be covered by a machine-learning rule first. Almost all information is machine-generated, including vast amounts of financial data for both business and personal use. To require all machine-generated evidence, which is routinely brought into court by both expert and lay witnesses, to undergo a thorough 702 reliability analysis could significantly delay trials and lead to unnecessary appeals. Alternatively, the Committee could decide that a rule on machine learning is needed now, but that capturing almost all machine-generated evidence is unnecessary. While some on the Committee may be concerned with an underinclusive rule, an overinclusive rule would have a greater impact on the courts. To the extent that the Committee concludes it will move forward with a rule at the conclusion of the comment period, it would be easier to strengthen or clarify a less-inclusive rule through amendment than it would be to repeal a rule that has overreached. 21 19 Psychologists are increasingly turning to tools powered by artificial intelligence (AI) to streamline their practice— about 1 in 10 use it at least monthly for note-taking and other administrative work. Barriers to Care in a Changing Practice Environment: 2024 Practitioner Pulse Survey, AM. PSYCH. ASS’N 1 (Dec. 2024). However, “many remain skeptical, with 71% reporting they’ve never used AI in their practice. Zara Abrams, Artificial Intelligence Is Reshaping How Psychologists Work, AM. PSYCH. ASS’N SERVS., INC. (last updated June 26, 2025). 20 Lawyers for Civ. Just., Comment Letter on Proposed Rule 707 on Machine-Generated Evidence, at 2 (Jan. 5, 2026) (“The new rule should be custom-made for its purpose, not a cross-reference to an existing rule. Rule 707 should not require each reader to interpolate the language of Rule 702(a)-(d), the vocabulary of human expert witnesses, into to the world of machines, models, and algorithms. Courts and lawyers will struggle with the linguistic mismatch.”). 21 Mandatory Rule 11 sanctions resulted in a satellite litigation without accomplishing its goal of deterring abuse. .” Fed. R. Civ. P. 11 advisory committee’s note to 1983 amendment (“Experience shows that in practice Rule 11 has not been effective in deterring abuses.”). Page 7 of 10 A. Narrow the rule to focus on machine learning At its November 2025 meeting, the Evidence Committee discussed an alternative to the proposed amendment on the “Output of a Process of Machine-Learning.” 22 This draft limited the rule by focusing on the specific concerns that have been brought to the attention of the Evidence Committee and are the impetus for moving a draft rule to public comment. Unlike the version of the rule published for public comment, this option provided a detailed definition of machine learning in the first sentence of the Committee Note, complete with descriptions of how these systems work and relate to artificial intelligence: Machine learning is an application of artificial intelligence that is characterized by providing systems the ability to automatically learn and improve on the basis of data or experience, without being explicitly programmed. Machine learning involves artificial intelligence systems that are used to perform complex tasks in a way that is similar to how humans solve problems. Machine-learning systems can make predictions or draw inferences from existing data supplied by humans. When a machine draws inferences and makes predictions, there are concerns about the reliability of that process, akin to the reliability concerns about expert witnesses. A definition would help ensure that parties and courts understand the scope and application of the rule, including that it addresses new technology and is not intended to slow down trials over machine outputs widely accepted by the public today and generally understood to be reliable. Further, this refinement would assist the court in assessing data where a human provides certain, often minimal, inputs, while the machine calculates other numbers or information. For example, truckers are responsible for starting their vehicle’s Electronic Logging Device (ELDs) at the start of their shift by logging in, but it is the device itself that automatically records driving time, miles driven, and other key data. 23 ELDs are widely accepted as reliable, but it would be unfair to burden a truck driver testifying about their injuries with a 702 inquiry into whether the device accurately recorded their inputs on the day of the crash. This draft proposal also has the added benefit of dropping the second sentence on simple scientific instruments, as it would not be necessary. The machine-learning alternative is also preferable to the alternative on “ComputerGenerated Evidence,” which the Evidence Committee also discussed. 24 The agenda book analysis is that “computer-generated is narrower than the “machine-generated” text provided by the proposed rule, it does not specifically address the concerns associated with the use of machine learning and artificial intelligence. Word processors, calculators, and digital cameras are all examples of simple computers that would all be considered reliable. 22 Capra Memo, supra note 6, at 141. 23 Truck drivers must input key data into the ELD, such as loading, fueling, off-duty, inspection, etc. to record what they are doing, but the ELD automatically tracks the day’s activities. About ELDs: Improving Safety Through Technology, FED. MOTOR CARRIER SAFETY ADMIN., DEP’T OF TRANSP. (last visited Feb. 12, 2026). 24 See Capra Memo, supra note 6, at 143. Page 8 of 10 B. Provide a definition to the rule’s application If the Committee decides to proceed with the current version of the amendment, it should provide a definition of “machine-generated” preferably in the text of the rule, but at a minimum, in the Committee Note. As drafted, there is no definition of “machine-generated” provided in either and the discussion of the proposed rule seems to indicate that it should be given a broad interpretation, 25 thus resulting in much of AAJ’s objection to the rule. Failing to provide a definition will result in courts applying the rule differently to the same instruments and technology. Circuit splits and intra-district splits could immediately develop. It would be better to limit the scope of the rule and be clear what the rule intends to cover. This will avoid second-guessing by parties and courts, uneven application of the rules, and satellite litigation. C. Exempt routinely used instruments The sentence to exclude simple scientific instruments would barely limit the number of items that judges would have to evaluate. By using both the word “simple” and scientific” in the caveat, the exception could even result in some parties questioning the reliability of date and timestamp data from mobile phones and other electronic devices. At a minimum, both the rule text and the Committee Note’s examples need to be crafted in a more sophisticated manner, as all these examples are items that a court could take judicial notice of as reliable under Rule 201. Should the Evidence Committee decide to move forward with a final rule, it would be beneficial to remove the second sentence. Alternatively, it may be clearer for both parties and courts to rewrite the second sentence with the aim of exempting routinely relied upon instruments: This rule does not apply to instruments routinely used to produce [generate] the output. The goal of the sentence is to create an exception for the output of instruments that are routinely used (and expected to be used) for the function being measured, captured, or produced. At its fall meeting, the Evidence Committee thought it should not cede coverage of the rule to the general public, which this sentence would not do, as the court would still be required to evaluate whether an instrument is routinely used. Machines that are routinely used but are not generally used for the function employed would not be exempted. The word “generate” would be preferable to the word “produce” only if the Evidence Committee adopts rule text other than “machinegenerated” evidence. The corresponding paragraph in Committee Note could be rewritten to address both the issues of routine use and some examples could look like this: 25 Id. at 143 (“‘Machine’ is defined as ‘an apparatus using or applying mechanical power and having several parts, each with a definite function and together performing a particular task.’ So that term covers everything from fax machines to bulldozers.”). Page 9 of 10 The final sentence of the rule is intended to give trial courts sufficient latitude to avoid unnecessary litigation over the output from simple scientific instruments that are routinely relied upon in everyday life. Examples might include the results of a phone log, geolocation data, or other metadata. Moreover, the rule does not apply when the court can take judicial notice that the machine output is reliable. See Rule 201. This exception also has the added benefit of functioning in tandem with instructions on judicial notice. V. Conclusion AAJ urges the Evidence Committee to reconsider what sort of rule is needed to address concerns regarding machine learning and AI and focus on a rule to specifically address those issues. The proposed rule applies to all sorts of machines that pose no reliability concerns and the exemption for simple scientific instruments is far too limited. The committee should pause and reevaluate how to draft a rule in light of the comments and testimony received. Thank you for considering these comments. Please direct any questions to Susan Steinman, Senior Director of Policy & Senior Counsel, at susan.steinman@justice.org. r Respectfully submitted, Bruce Plaxen President American Association for Justice Page 10 of 10 An official website of the United States Government. < Back to Document Comments (/document/USC-RULES-EV-2025-0034-0001/comment) Share PUBLIC SUBMISSION Comment from Center for Democracy & Technology and Five Other Civl Rights, Civil Liberties and Professional Organizations Posted by the United States Courts on Feb 17, 2026 Docket (/docket/USC-RULES-EV-2025-0034) / Document (USC-RULES-EV-2025-0034-0001) (/document/USC-RULES-EV-2025-0034-0001) / Comment Comment Attachments 1 Evidence Rule 707 Joint Comment Final Download (https://downloads.regulations.gov/USC-RULES-EV-2025-0034-0058/attachment_1.pdf) Give Feedback Civil liberties, civil rights, and professional organizations are asking the Advisory Committee to reconsider proposed FRE 707. It would apply to AI-generated evidence the rules of admissibility that currently apply to expert witness testimony in existing rule of evidence, FRE 702. We point out that the appropriate criteria for assessing the reliability of AI-generated evidence differ from the criteria in FRE 702 for assessing the reliability of an expert witness’s testimony, making this an improper fit. The views of these organizations, ACM U.S. Technology Policy Committee, Asian Americans Advancing Justice | AAJC, Center for Democracy & Technology, Electronic Privacy Information Center (EPIC), Fight for the Future, and UnidosUS are more fully set forth in their full Comment, which is attached. Comment ID USC-RULES-EV-2025-0034-0058 Tracking Number mlp-mz17-ub8g Comment Details Submitter Info Received Date Feb 16, 2026 This agency received 6 duplicate or significantly similar comments. About Bulk Data Download Learn Reports FAQ Commenting Guidance (/bulkdownload) (/agencies) (/learn) (/dotreports) (/faq) (/commenting-guidance) Privacy & Security Notice (/privacy-notice) | User Notice (/user-notice) | Accessibility Statement (/accessibility) | API Requests (https://open.gsa.gov/api/regulationsgov/) | FOIA (https://www.gsa.gov/reference/freedom-of-information-act-foia) Support (/support) Give Feedback (/about) Agencies Civil Rights, Civil Liberties, and Professional Organizations’ Joint Comment on Proposed Federal Rule of Evidence 707 Concerning Admissibility of Machine-Generated Evidence February 16, 2026 We are civil liberties, civil rights, and professional organizations that have engaged in policy debates surrounding the use of artificial intelligence (AI) in various contexts. We are committed to ensuring that AI use does not exacerbate societal bias and civil liberties threats. The Advisory Committee has proposed a Federal Rule of Evidence 707 that would govern the admissibility in federal courts of machine-generated information (which includes AI-generated information) based on the reliability of that information. Courts’ decisions to admit or exclude AIgenerated evidence can result in wrongful incarceration as well as significant liability. For example, use of AI-powered facial recognition systems has resulted in a number of wrongful arrests. This is a high-risk use of AI, and demands exacting scrutiny. The proposed rule would apply to AI-generated evidence the rules of admissibility that currently apply to expert witness testimony in an existing rule of evidence, FRE 702. But the appropriate criteria for assessing the reliability of AI-generated evidence differ from the criteria for assessing the reliability of an expert witness’s testimony, making this an improper fit. For example, AIgenerated information is more reliable if the system that produced it was trained on unbiased data of high quality. Nothing in Rule 702 assesses the training data, which instead focuses on the knowledge and experience of the expert. A rule governing the admissibility of AI-generated information in court proceedings is needed, and the Advisory Committee, to its credit, has taken up this challenging task. While it has held hearings and received testimony from a variety of witnesses with legal training, it has not yet heard from certain technical experts and people steeped in AI policy-making who could help it fashion a properly-focused rule. We urge the Committee to reconsider the text of the proposed rule and to reach out to solicit additional input from a broader range of stakeholders. We would welcome the opportunity to more fully participate in its efforts to develop a new rule of evidence focused on AI-generated information. Sincerely ACM U.S. Technology Policy Committee Asian Americans Advancing Justice | AAJC Center for Democracy & Technology Electronic Privacy Information Center (EPIC) Fight for the Future UnidosUS An official website of the United States Government. < Back to Document Comments (/document/USC-RULES-EV-2025-0034-0001/comment) Share PUBLIC SUBMISSION Comment from Coalition for Prior Conviction Impeachment Reform Posted by the United States Courts on Feb 17, 2026 Docket (/docket/USC-RULES-EV-2025-0034) / Document (USC-RULES-EV-2025-0034-0001) (/document/USC-RULES-EV-2025-0034-0001) / Comment Comment Please find attached a comment on the proposed amendment to Rule 609. Attachments 1 Comment on Rule 609 Amendment from Coalition for Prior Conviction Impeachment Reform Download (https://downloads.regulations.gov/USC-RULES-EV-2025-0034-0059/attachment_1.pdf) Give Feedback [ Comment ID USC-RULES-EV-2025-0034-0059 Tracking Number mlp-npbd-f83z Comment Details Submitter Info UCONN I SCHOOLOFLAW Julia Simon-Kerr Evangeline Starr Professor of Law Hon. Jesse Furman Chair, Advisory Committee on the Rules of Evidence Administrative Office of the United States One Columbus Circle, NE Washington, DC 20544 February 16, 2026 Dear Judge Furman and Advisory Committee Members, We write on behalf of the Coalition for Prior Conviction Impeachment Reform, a group of eleven professors who study Federal Rule of Evidence 609 and are convinced of the need to reform it.1 We know from our research and efforts that change is hard to achieve in the prior conviction impeachment sphere. We therefore express our appreciation for the fact that after years of work by the Reporter, Academic Liaison, Committee, and invited experts (including Coalition member Jeffrey Bellin) a proposal has reached this stage. We write to support the proposed reform and to make two suggestions. We have voiced our support before, by means of a letter dated September 30, 2024, and comments (written and oral) on our behalf from Coalition member Professor John Blume. We will distill our main points under the headings of the two goals of the Federal Rules of Evidence laid out in Rule 102: ascertaining truth and securing just determinations.2 The proposed amendment has the potential to limit the extent to which Federal Rule of Evidence 609(a)(1)(B) detracts from ascertaining the truth. As the members of our Coalition, among many others, have long argued,3 this rule is antithetical to the truth-seeking mission of the Federal Rules of Evidence for a host of reasons. For example, as we have described in previous submissions, there is no evidence that prior convictions admitted under 609(a)(1)(B) have 1 Professors Jeffrey Bellin (Vanderbilt Law School), John Blume (Cornell Law School), Bennett Capers (Fordham University School of Law), Montré Carodine (University of Alabama School of Law), Jasmine Gonzales Rose (Boston University School of Law), Lisa Kern Griffin (Duke University School of Law), John D. King (Washington and Lee School of Law), Colin Miller (University of South Carolina School of Law), Aviva Orenstein (Indiana University Maurer School of Law), Anna Roberts (Brooklyn Law School), and Julia Simon-Kerr (The University of Connecticut School of Law). 2 See Daniel J. Capra & Liesa L. Richter, Poetry in Motion: The Federal Rules of Evidence and Forward Progress as an Imperative, 99 B.U. L. Rev. 1873, 1879 (2019) (“[T]he goal of the Evidence Rules [is] to ascertain truth and secure just determinations”). 3 See, e.g., Note, The Evidentiary Use of Constitutionally Defective Prior Convictions, 68 Colum. L. Rev. 1168, 1171 (1968) (describing as “widely recognized” the fact “that informing a jury of a prior conviction before they retire to reach a verdict may lead to a less accurate factual determination,” particularly if it “deters the defendant from testifying in his own behalf.”) probative value on the question of a witness’s truthfulness.4 Social science data suggests, to the contrary, that such prior convictions likely will not assist fact-finders in predicting a defendant’s truthfulness.5 Yet, courts routinely admit prior convictions under the existing balancing test, imagining against the dictates of both empirical evidence and logic that their probative value on truthfulness outweighs the known and documented risk of unfair prejudice.6 Cases ending in “wrongful convictions,” such as those studied by John Blume in his research into the effects of prior conviction impeachment, bear out the causal relationship between the rule and catastrophic failures in truth-seeking.7 Adding “substantially” to the balancing test will promote truth-seeking if it corrects these misapplications and restrains judges from admitting prior convictions against the accused. Justice obviously is not served when rules of evidence point away from truth-seeking. But there are additional ways in which Federal Rule of Evidence 609(a)(1)(B) detracts from the likelihood of a just determination. These problems too, may be limited by the proposed amendment. First, the American legal system is committed to procedural justice, a commitment that includes a meaningful opportunity to speak.8 Yet the threat of prior conviction impeachment contributes to the silence of the accused, including those with stories of innocence to tell.9 Those who exercise their constitutional right to remain silent risk being penalized by jurors, despite instructions to the contrary.10 Further, as has been detailed in many submissions, racial injustices are compounded by prior conviction impeachment.11 And finally, prior conviction impeachment 4 See, e.g., Anna Roberts & Julia Simon-Kerr, Reforming Prior Conviction Impeachment, 50 Fordham Urban L.J. 377, 384-90 (2023). 5 See, e.g., id. at 386-88 (2023). 6 See, e.g., Theodore Eisenberg & Valerie P. Hans, Taking a Stand on Taking the Stand: The Effect of a Prior Criminal Record on the Decision to Testify and on Trial Outcomes, 94 Cornell L. Rev. 1353, 1359–61 (2009); Jeffrey Bellin, Circumventing Congress: How the Federal Courts Opened the Door to Impeaching Criminal Defendants with Prior Convictions, 42 U.C. Davis L. Rev. 289, 325–26 (2008); Anna Roberts, Reclaiming the Importance of the Defendant's Testimony: Prior Conviction Impeachment and the Fight Against Implicit Stereotyping, 83 U. Chi. L. Rev. 835, 864 (2016). 7 John H. Blume, The Dilemma of the Criminal Defendant with a Prior Record—Lessons from the Wrongfully Convicted, 5 J. Empirical Legal Stud. 477, 491 (2008). 8 See Michael M. O’Hear, Plea Bargaining and Procedural Justice, 42 Ga. L. Rev. 407, 466 (2008) (stating that “voice [is] often viewed as the core attribute of procedural justice.”). 9 See, e.g., Jeffrey Bellin, The Silence Penalty, 103 Iowa L. Rev. 395, 432–33 (2018); John H. Blume, The Dilemma of the Criminal Defendant with a Prior Record—Lessons from the Wrongfully Convicted, 5 J. Empirical Legal Stud. 477, 491 (2008). See also Theodore Eisenberg & Valerie P. Hans, Taking a Stand on Taking the Stand: The Effect of a Prior Criminal Record on the Decision to Testify and on Trial Outcomes, 94 Cornell L. Rev. 1353, 1370 (2009) (“In the cases in which defendants testified, judges reported that, on average, defendant testimony was more important than that of the police, of informants, of codefendants, and of expert witnesses.”); Alexandra Natapoff, Speechless: The Silencing of Criminal Defendants, 80 NYU L. Rev. 1449, 1459–60 (2005) (“Defendants do not testify largely because it is so dangerous. . . It . . . allows the government to elicit the defendant’s criminal history . . . which may dissuade the jury from hearing the substance of the defendant’s story, from having sympathy with the defendant, or from disbelieving the government.”). 10 See, e.g., Jeffrey Bellin, The Silence Penalty, 103 Iowa L. Rev. 395, 412–15 (2018). 11 See, e.g., Montré Carodine, “The Mis-Characterization of the Negro”: A Race Critique of the Prior Conviction Impeachment Rule, 84 Ind. L.J. 521, 549 (2009) (noting that “one must keep in mind that most people at that time-as 2 evidence is predictably misused by jurors, despite instructions to the contrary.12 For example, prior conviction impeachment lowers the burden of proof in close cases, denying the accused the benefit of the beyond a reasonable doubt standard.13 It is our belief that the proposed amendment, by limiting the impeachment of the accused in criminal cases, will go some way towards aligning the Federal Rules of Evidence with their purported goal of promoting just determinations. In sum, the Coalition believes that this proposed rule is a step in the right direction.14 In light of our research, the Committee’s discussions, and the Committee’s policy of tackling more than one issue with a rule at the same time where possible,15 we make two suggestions: Suggestion 1 The Reporter has detailed numerous examples of trial courts failing to honor the Federal Rule of Evidence 609(a)(1)(B) balancing test. The record also suggests that a cause of those failures is that trial judges lack incentives to conduct careful balancing.16 A central cause of this incentive problem is judicially-imposed restrictions on the power to appeal an adverse ruling admitting convictions under 609(a)(1)(B).17 Under Luce, the accused must testify in order to appeal such a ruling. Post-Luce, Rule 609 appeals have “plummeted.”18 Under Ohler, the accused cannot appeal such a ruling if he testifies about his prior convictions on direct. Without the ability to introduce the convictions himself, the accused is unable to be candid with jurors about them, or to frame or contextualize them. To preserve the right to appeal the trial judge’s admissibility decision under Federal Rule of Evidence 609(a)(1)(B), the accused must instead wait for the evidence to be raised by the prosecutor on cross, thus risking the inevitable jury inference that he was hiding the evidence. The structural barriers to prior conviction impeachment appeals brought about by Luce and Ohler have created a vacuum of appellate regulation of trial courts as they conduct 609 is true today-saw a Black face when they thought about the criminal element in society.”); Anna Roberts & Julia Simon-Kerr, Reforming Prior Conviction Impeachment, 50 Fordham Urban L.J. 377, 392-95 (2023). 12 See, e.g., Theodore Eisenberg & Valerie P. Hans, Taking a Stand on Taking the Stand: The Effect of a Prior Criminal Record on the Decision to Testify and on Trial Outcomes, 94 Cornell L. Rev. 1353, 1359–61 (2009). 13 Id. 14 See Email from John H. Blume to Rules Committee Secretary (Jan. 1, 2026), Advisory Comm. on Evidence Rules, Hearing on Proposed Amendments 4 (Jan. 15, 2026), https://www.uscourts.gov/sites/default/files/document/jan-15-hearing-schedule-and-testimony-packet-final.pdf. Our Coalition’s concerns about Rule 609 extend beyond the Rule 609(a)(1)(B) balancing test, and beyond the two suggestions that we make, and we therefore hope that greater change will follow. 15 See Memorandum from Daniel J. Capra, Reporter, to Advisory Committee on Evidence Rules 37 (Apr. 1, 2025) (“[T]he Committee’s policy has always been that if a rule is going to be amended, there might be improvements that can—and should—be made even though those improvements are not enough to justify an amendment standing alone. . . [I]f the Committee is going to amend a rule, that is a good time to make it the best it can be.”). 16 See id. at 17 (mentioning a Committee member’s argument that “the problem was not the rule, but that trial courts are not incentivized to apply it correctly because there is no review over Rule 609 decisions to admit evidence.”). 17 See id. 18 See id. at 24 n.11. 3 balancing. Insulated from review,19 trial courts have used their discretion to misapply the balancing test and created the urgent need for the current proposed amendment. If the proposed balancing test is to ameliorate the problems it seeks to address, these impediments to appellate review should be tackled now.20 At least seventeen states have declined to follow Ohler.21 At least eleven states have declined to adopt the Luce requirement that in order to preserve the right to appeal a prior conviction impeachment ruling the accused must testify.22 While most of those eleven states developed that stance through case law, Tennessee offers an example of a state that specifies in its version of Rule 609 that a decision not to testify does not destroy the ability to appeal a prior conviction impeachment ruling: If the court makes a final determination that [convictions are] admissible for impeachment purposes, the accused need not actually testify at the trial to later challenge the propriety of the determination.23 19 See Erin R. Collins, Evidence Rules for Decarceration, 50 Fordham Urb. L.J. 353, 364 (2023) (“FRE 609 decisions, no matter how erroneous or unfair, are effectively immunized from appellate correction in many cases.”). 20 Although the Reporter has stated that “[a]brogating Luce is a possibility that will be explored at future meetings if the current proposal to amend Rule 609 is not approved,” tackling Luce is important to the success of the current proposal. Memorandum from Daniel J. Capra, Reporter, to Advisory Committee on Evidence Rules 18 (Apr. 1, 2025). 21 See People v. Carpenter, 988 P.2d 531, 556 (Cal. 1999); McGill v. DIA Airport Parking, LLC, 395 P.3d 1153, 1156–57 (Colo. App. 2016); State v. Daly, 623 N.W.2d 799, 800–01 (Iowa 2001); Cure v. State, 26 A.3d 899, 911– 12 (Md. 2011); State v. Swanson, 707 N.W.2d 645, 654 (Minn. 2006); McGee v. State, 569 So.2d 1191, 1194–95 (Miss. 1990), overruled on other grounds by White v. State, 785 So.2d 1059, 1061 (Miss. 2001); Malone v. State, 829 So.2d 1253, 1259–60 (Miss. Ct. App. 2002); Pineda v. State, 120 Nev. 204, 208–10 (2004); Whisler v. State, 121 Nev. 401, 406 (2005); Zola v. Kelley, 149 N.H. 648, 826 A.2d 589, 591–93 (2003); State v. Allen, 323 P.3d 925, 928–30 (N.M. Ct. App. 2013); State v. Ross, 329 N.C. 108, 405 S.E.2d 158, 163–64 (1991) (“A defendant would face an unfair dilemma if forced to choose between devastating cross-examination about a conviction and waiver of his right to appeal the denial of a pretrial motion”); State v. Phillips, 298 Or. App. 743, 450 P.3d 54, 55–56 (2019); Comm. v. Stevenson, 318 A.3d 1264, 1276–82 (Pa. 2024) (collecting cases and stating that “[w]hile a handful of jurisdictions have aligned with the Ohler majority’s rule, a majority of them have adopted the reasoning of the Ohler dissent.”); State v. Mueller, 319 S.C. 266, 460 S.E.2d 409, 411 (1995); State v. Keiser, 174 Vt. 87, 807 A.2d 378, 388 (2002); State v. Thang, 145 Wash.2d 630, 41 P.3d 1159, 1168 (2002) (“We agree with Justice Souter's analysis [in dissent]. A defense lawyer who introduces preemptive testimony only after losing a battle to exclude it cannot be said to introduce the evidence voluntarily.”); State v. Gary M.B., 270 Wis.2d 62, 676 N.W.2d 475, 480–83 (2004). 22 State v. Sineros, 137 Ariz. 323, 325 (1983); Commonwealth v. Crouse, 447 Mass. 558, 564 (2006); People v. McBride, 413 Mich. 341, 345 (1982); People v. Frey, 168 Mich. App. 310, 317 (1988); State v. Jones, 271 N.W.2d 534, 537 (Minn. 1978); Comm. Comment to Minn. R. Evid. 609(a) (“Contrary to the practice in federal courts, the defendant can preserve the issue at a motion in limine and need not testify to litigate the issue in post trial motions and appeals.”); State v. Swanson, 707 N.W.2d 645, 654 (Minn. 2006); Hickson v. State, 697 So.2d 391, 396–98 (Miss. 1997); Warren v. State, 121 Nev. 886, 894–95 (2005) (finding offer of proof sufficient); State v. Whitehead, 517 A.2d 373, 376–77 (N.J. 1986); People v. Contreras, 485 N.Y.S.2d 261, 263 (N.Y. App. Div. 1985); State v. Eugene, 340 N.W.2d 18, 29 (N.D. 1983) (“[a] defendant does not waive his objection to an adverse ruling on a motion in limine by introducing his prior convictions on direct examination.”); State v. McClure, 298 Or. 336, 342 n.4 (1984); Commonwealth v. Richardson, 347 Pa. Super. 564, 569–71 (1985); Tenn. R. Evid. 609(a)(3). 23 Tenn. R. Evid. 609(a)(3). 4 We urge the Committee to consider its ability to take steps to address these barriers to appeal and the improper implementation of carefully crafted balancing tests that results. Suggestion 2 When 609(a)(1)(B) is misapplied, the most obviously responsible party is the judge. The proposed amendment targets that concern, sending a “signal” to judges.24 But every judicial decision to admit convictions that are similar to the charge at hand, needlessly cumulative, inflammatory, or otherwise inappropriate, is preceded by a prosecutorial decision to proffer at least that many convictions. As the Reporter puts it, sometimes the prosecution gets “greedy.”25 The Committee might, in the spirit of addressing more than one issue at once,26 usefully send a signal on that front too. In 2015, Rule of Civil Procedure 1 was amended “to emphasize that just as the court should construe and administer these rules to secure the just, speedy, and inexpensive determination of every action, so the parties share the responsibility to employ the rules in the same way.”27 By analogy, if the courts have an obligation to strive for truth and justice in their application of the rules of evidence, so too do prosecutors. Prosecutors, indeed, have both an ethical and constitutional duty to do justice, in evidentiary as in other spheres.28 Some have called for prosecutors to exercise restraint in the proffering of this form of evidence,29 whose racial disparity is so stark,30 and whose misuse so likely.31 24 See Memorandum from Daniel J. Capra, Reporter, to Advisory Committee on Evidence Rules 17 (Oct. 1, 2024) (“The argument in favor of the amendment is that a slight change to the balancing test can be a signal to courts that they need to more carefully weigh prejudicial effect and probative value, and give defendants the protection that Congress intended.”). 25 See Liesa Richter, District Court Rulings on Rule 609(a)(1)(B) Impeachment—2009–present 30 (case digest, updated and with commentary by Daniel J. Capra) (Apr. 1, 2005) (“Comment: This is just a case in which the government was greedy. They were already going to impeach the defendant with six automatically admissible convictions. And yet they wanted to also impeach with a conviction that was similar to the crime charged. In these circumstances, the argument that the conviction is necessary for, and will be limited to, impeachment, seems disingenuous.”) (referring to United States v. Cunningham, 2012 WL 12865641 (W.D. Mich. 2012). 26 See supra note 15. 27 F.R.C.P. 1, Advisory Committee Note to the 2015 Amendments (emphasis added). 28 See Daniel J. Capra & Liesa L. Richter, Poetry in Motion: The Federal Rules of Evidence and Forward Progress as an Imperative, 99 B.U. L. Rev. 1873, 1911 (2019) (“[T]he DOJ is, of course, charged with pursuing ‘justice’. . . .”). 29 See, e.g., Steven Zeidman, Some Modest Proposals for a Progressive Prosecutor, 5 UCLA CRIM. JUST. L. REV. 23, 43 (2021) (recommending prosecutorial refusal to engage in this practice). 30 See ABA CRIM. JUST. STANDARDS FOR THE PROSECUTION FUNCTION § 3-1.6(b) (4th ed. 2017) (“A prosecutor’s office should be proactive in efforts to detect, investigate, and eliminate improper biases, with particular attention to historically persistent biases like race, in all of its work. A prosecutor’s office should regularly assess the potential for biased or unfairly disparate impacts of its policies on communities within the prosecutor’s jurisdiction, and eliminate those impacts that cannot be properly justified.”); Anna Roberts & Julia Simon-Kerr, Reforming Prior Conviction Impeachment, 50 Fordham Urban L.J. 377, 392–95 (2023). 31 See, e.g., Theodore Eisenberg & Valerie P. Hans, Taking a Stand on Taking the Stand: The Effect of a Prior Criminal Record on the Decision to Testify and on Trial Outcomes, 94 Cornell L. Rev. 1353, 1359–61 (2009). 5 The Coalition thus suggests that the Committee include within the Committee Note a reminder that prosecutors have a duty to do justice in this sphere. Sincerely, Professors Julia Simon-Kerr & Anna Roberts On Behalf of The Coalition for Prior Conviction Impeachment Reform 6 An official website of the United States Government. Back to Document Comments (/document/USC-RULES-EV-2025-0034-0001/comment) Share PUBLIC SUBMISSION Comment from Remy, Joseph Posted by the United States Courts on Feb 17, 2026 Docket (/docket/USC-RULES-EV-2025-0034) / Document (USC-RULES-EV-2025-0034-0001) (/document/USC-RULES-EV-2025-0034-0001) / Comment Comment I respectfully submit this comment in opposition to proposed Federal Rule of Evidence 707. While the Rule aims to address challenges presented by AI‑generated and other machine‑generated evidence, the proposal is unnecessary, impractical, and risks undermining the flexibility and fairness of the existing evidentiary framework. What is more, despite comments to the contrary, both plaintiffs’ attorneys and in‑house counsel—which rarely align—oppose the Rule because courts have not struggled to manage machine‑generated outputs under existing evidentiary rules. The proposal addresses a problem that’s not a problem while ignoring the very real challenges facing the judiciary. Indeed, Federal Rules of Evidence 702, 901, 902, and 403 already provide a robust structure for evaluating expert‑like evidence, authenticating digital outputs, and excluding unreliable or prejudicial material. Legal practitioners analyzing the proposal observe that courts are already capable of handling these issues case by case, and that adopting a new rule now would be premature given the limited case law involving AI‑generated evidence. Due to the present state of federal and state budgets, Rule 707 also risks triggering unnecessary and expensive satellite litigation over whether routine digital outputs require a full Rule 702 foundation. As one analysis warns, the proposal could lead to “costly battles” over simple software printouts, turning ordinary evidentiary questions into Daubert‑style disputes even when the output consists merely of raw data. This risk of overreach would slow litigation and burden courts and parties alike. Give Feedback As recognized by both the National Institute of Standards and Technology as well as the International Association of Chiefs of Police, the use of AI to produce an output that would carry criminal and/or civil liability should be reviewed by a living person; who then adopts it. This living person would be subject to cross examination and any output subject to the existing rules of admissibility. Finally, the enactment of any such rule should be preceded with sufficient training to judges on artificial intelligence, a condition not sufficiently met by federal and subsequent states that may enact such a rule. Accordingly, the proposed Rule 707 is unnecessary, overly broad, and potentially harmful to both litigants and the judicial system. Existing rules already provide sufficient safeguards. I urge the Committee to decline adoption of Rule 707 and continue relying on the well‑established evidentiary framework that has long served the courts. Comment ID USC-RULES-EV-2025-0034-0060 Tracking Number mlp-nthn-woiv Comment Details Submitter Info Received Date About Bulk Data Download (/about) Agencies Learn Reports FAQ Give Feedback Feb 16, 2026 Commenting Guidance (/bulkdownload) (/agencies) (/learn) (/dotreports) (/faq) (/commenting-guidance) Privacy & Security Notice (/privacy-notice) | User Notice (/user-notice) | Accessibility Statement (/accessibility) | API Requests (https://open.gsa.gov/api/regulationsgov/) | FOIA (https://www.gsa.gov/reference/freedom-of-information-act-foia) Support (/support) An official website of the United States Government. < Back to Document Comments (/document/USC-RULES-EV-2025-0034-0001/comment) Share PUBLIC SUBMISSION Comment from Redgrave, Jonathan Posted by the United States Courts on Feb 17, 2026 Docket (/docket/USC-RULES-EV-2025-0034) / Document (USC-RULES-EV-2025-0034-0001) (/document/USC-RULES-EV-2025-0034-0001) / Comment Comment See attached file(s) [ Attachments 1 2026.02.16 Redgrave Comment on Proposed Federal Rule of Evidence 707 Give Feedback Download (https://downloads.regulations.gov/USC-RULES-EV-2025-0034-0061/attachment_1.pdf) Comment ID USC-RULES-EV-2025-0034-0061 Tracking Number mlp-p0in-lqvt Comment Details Received Date Submitter Info tl ~I?Q!{AVE LLP Jonathan Redgrave Benjamin Redgrave Timothy Redgrave Partner Redgrave LLP 4800 Westfield Blvd. Suite 250 Chantilly, VA 20151 202.603.1497 jredgrave@redgravellp.com Counsel Redgrave LLP 230 West Monroe St. Suite 210 Chicago, IL 60606 312.758.5887 bredgrave@redgravellp.com 6201 McQueen Dr. Durham, NC 27705 202.505.0897 tpredgrave@gmail.com February 16, 2026 Advisory Committee on Evidence Rules U.S. Judicial Conference Committee on Rules of Practice and Procedure Thurgood Marshall Federal Judiciary Building 1 Columbus Circle, NE Washington D.C. 20002 Re: Comments on Proposed New Fed. R. Evid. 707: Why A Modified Approach to AI-Generated Evidence is Needed Dear Advisory Committee members: The undersigned 1 respectfully submit the following comments and suggestions regarding the proposed introduction of a new federal rule of evidence (“Rule 707”) to address certain AIgenerated evidence (euphemistically called “machine-generated evidence”). The comments supplement the oral testimony from Jonathan Redgrave, at the January 29, 2026, Advisory Committee hearing. Proposed Rule 707 represents a well-intentioned but flawed attempt to address the admissibility of artificial intelligence-generated (“AI-generated”) content in federal court proceedings. While the drafters sought to create a framework to guide judges and parties with respect to evaluating and admitting AI-generated evidence, the rule as currently constructed fails to achieve its stated objectives and instead creates a dangerous equivalence between AI outputs and human expert testimony under Rule 702. As set forth in Section III, this equivalence is inappropriate, scientifically unsound, and risks undermining the reliability standards that have long 1 Jonathan Redgrave is a Partner at Redgrave LLP and has extensive involvement with litigation issues over the past thirty years in federal and state courts across the country. He was instrumental in the foundation of The Sedona Conference’s Working Group Series, including those addressing electronic discovery and information governance. He regularly counsels clients regarding issues involved in the application of artificial intelligence systems in business and the impacts on dispute resolution. Benjamin Redgrave is a Counsel at Redgrave LLP where his practice focuses on electronic discovery and evidence issues, including legal issues surrounding the cutting-edge application of artificial intelligence systems in business. Timothy Redgrave is a recent graduate from the University of Notre Dame, receiving a Masters Degree in Computer Science with a specialization in artificial intelligence systems. Highlights of his studies include interrogating the efficacy and stability of adversarial attacks against machine learning systems under varying batch sizes (https://doi.org/10.1109/CVPRW59228.2023.00235) and creating a fully invertible prototypical neural network architecture for accurate and explainable predictive and generative modeling. (https://doi.org/10.1007/9783-031-72913-3_13). The views expressed by the authors are theirs alone and do not necessarily represent the views of any firms, employers, or organizations to which they belong, or any clients they represent. J. Redgrave, B. Redgrave, and T. Redgrave Comment on Proposed Rule 707 February 16, 2026 Page 2 protected the integrity of expert evidence in federal courts. Consequently, the Advisory Committee should reject the current proposed rule. At the same time, the undersigned recognize and agree with the Advisory Committee and a minority of the comments that say that there is a need for action. Without additional guidance, we respectfully predict that courts will be confronted with a morass of potential uses of AI-generated outputs that either directly or indirectly impact expert witness testimony. This, in turn, will create a situation where conflicting approaches adopted in good faith by different judges in different jurisdictions will emerge and result in drastically different outcomes based on the admission or exclusion of AI-generated outputs. Such divergence will negatively impact the predictability of proceedings, destabilizing the basic tenets of fairness and justice that undergird the criminal and civil rules of procedure. 2 Fortunately, we believe that a viable alternative approach exists. Under this alternative framework, AI would be treated as a tool, not a witness, while remaining admissible as fact evidence when proper foundation is established pursuant to the existing rules. As such, in addition to rejecting the current rule, the Advisory Committee should draft a new one that: (1) prohibits AI outputs from serving as expert testimony altogether; (2) recognizes that experts may appropriately rely on AI tools in forming their opinions, provided they can establish the reliability of that reliance under existing Rule 702 factors; and (3) acknowledges that AI-generated content may be admissible as ordinary fact evidence when relevant and properly authenticated under Rules 401, 403, and 901. Our proposed alternative to the current rule that would accomplish these goals is set forth for consideration in Section II and the reasoning behind our proposal is set forth in Section IV. Finally, our view on other proposed rules addressing AI-generated outputs and evidence can be found in Section V. I. The Need to Pause for Reflection and Refinement By our count, a significant majority of commentators have urged the Advisory Committee to slow down and refine the proposed rule before adoption. 3 Common themes among those urging delay include: 2 • The alleged problem may not actually exist (no cases found where AI was offered without an expert); • Existing rules (702, 901, 902) may be adequate; See, e.g., Submission from Washington Legal Foundation (https://www.regulations.gov/comment/USC-RULES-EV2025-0034-0042). 3 See, e.g., Submission from Lawyers for Civil Justice, at 10 (https://www.regulations.gov/comment/USC-RULESEV-2025-0034-0013) (arguing that “the need for an appropriate rule vastly outweighs the utility of an immediate rule” and that “an incomplete or inadequate rule is certain to cause more harm than allowing courts to address emerging issues as the Advisory Committee works to refine its proposal”). J. Redgrave, B. Redgrave, and T. Redgrave Comment on Proposed Rule 707 February 16, 2026 Page 3 • The rule is overly broad and vague; • Technology is changing too rapidly; • More judicial experience is needed before codifying; and • The risk of unintended consequences. The Advisory Committee itself acknowledged “the limits of its expertise on matters of technology and deemed public comment as the best way to obtain the necessary information to support or reject the rule.” Advisory Committee on Rules of Practice and Procedure, Agenda Book at 59 (June 10, 2025) (explaining that the Advisory Committee expects to “receive critically important information during the public comment period about the need for this new rule”). At the same time, certain commentators endorsed the work of the Advisory Committee and suggest the rule should move forward. 4 Many of the commentators in this cohort noted that judges are already equipped with significant guidance and discretion under the existing evidence rules, including the recently amended Fed. R. Evid. 702, to appropriately manage issues that may arise with respect to computer-generated evidence. These commentators specifically endorse the Advisory Committee’s use of Rule 702 as the anchor for the new rule. While the outcome in rulemaking should not be driven by a “popular vote,” the contrast in opinions and submissions is stark. One of the undersigned, Jonathan Redgrave, has been an observer or participant in the federal judiciary’s rulemaking processes (primarily proposed civil and evidence rules) for more than 25 years and cannot recall a time when the divergence in views on a proposed rule was so significant irrespective of traditional advocacy positions (e.g., typical “plaintiffs” v. “defendants,” “patient” v “provider,” or “employees” v. “employers”). In addition, the groundswell of comments towards the end of the comment period suggests to us that AIgenerated evidentiary issues and proposed Rule 707 itself have only garnered significant attention outside of the Advisory Committee in the later stages of the rulemaking process to date. While that reality may be unfortunate (and potentially frustrating) given the significant undertaking by the Advisory Committee to draft and publish a proposed rule, we urge the Advisory Committee to step back with enough space to consider and address the comments that are being submitted from various viewpoints across the litigation spectrum (criminal and civil) and obtain more feedback on alternative rule proposals—though the reset in the rulemaking process need not be elongated or abandoned. 4 Paradoxically, some of these same commentators also suggested modifications that likely would require republication under the rulemaking process. J. Redgrave, B. Redgrave, and T. Redgrave Comment on Proposed Rule 707 February 16, 2026 Page 4 II. Proposed Alternative Federal Rule of Evidence 707: AI-Generated Evidence 5 Consistent with the foregoing plea for caution, we believe the alternative rule set forth below is a workable framework that could be vetted with republication of the proposed rule within the next rulemaking cycle, and we urge the Advisory Committee to do so. Rule 707: Expert Testimony Considering or Relying on Artificial Intelligence System Outputs (a) Definitions. For purposes of this rule, “Artificial Intelligence System” refers to any computational model, machine learning algorithm, or similar automated tool or process that produces outputs based on pattern recognition, prediction, or data analysis, and as it comes from a machine and not a person, is not capable of taking an oath, being cross-examined, or providing specialized knowledge in the manner contemplated by Rule 702. (b) Expert Testimony by Artificial Intelligence Systems Prohibited. No outputs generated by an Artificial Intelligence System shall be admitted as expert testimony under Rule 702 or otherwise. Artificial Intelligence System outputs may not serve as substitutes for human expert witnesses and may not be offered as opinions requiring specialized knowledge, skill, experience, training, or education. (c) Expert Use of Artificial Intelligence Systems Permitted. An expert witness may consider or rely on outputs of Artificial Intelligence Systems in forming their opinions, provided that: (1) the expert independently satisfies the requirements of Rule 702(a)–(d) with respect to their own testimony; (2) the expert demonstrates the reliability of the Artificial Intelligence System for the specific task, including providing appropriate foundation, qualified by knowledge, training, and experience, for the reliability, transparency, interpretability, and validation of the Artificial Intelligence System and its outputs generally and as applied to the facts of the case; and (3) the expert is able to explain their methodology, reasoning, and the factual basis for their consideration and/or reliance on the Artificial 5 While we acknowledge that there is a great divergence in various alternatives proposed by different commentators and the Advisory Committee itself, we have developed a rubric that draws inspiration primarily (but not exclusively) from the Advisory Committee’s prior work, the oral testimony of Prof. Roth (January 29, 2026) and her accompanying January 15, 2026 letter summary of her testimony, the submission of the Federal Courts Committee of the New York City Bar Association (https://www.regulations.gov/comment/USC-RULES-EV-2025-0034-0046), and the submission of Lawyers for Civil Justice (https://www.regulations.gov/comment/USC-RULES-EV-2025-0034-0013). J. Redgrave, B. Redgrave, and T. Redgrave Comment on Proposed Rule 707 February 16, 2026 Page 5 Intelligence System. (d) Artificial Intelligence System Outputs as Fact Evidence. Artificial Intelligence System outputs may be admitted as ordinary fact evidence when relevant and properly authenticated under Rules 401, 403, and 901. The party offering such evidence must establish its authenticity and relevance, and the court must ensure that the probative value is not substantially outweighed by any risk of unfair prejudice, confusion, or misleading the jury. Depending on the proposed use of the Artificial Intelligence System Outputs, the foundation may necessitate human testimony, qualified by knowledge, training, and experience, for the reliability, transparency, interpretability, and validation of the Artificial Intelligence System and its outputs generally and as applied to the facts of the case. In such circumstances, the testimony does not convert the output into expert testimony that is prohibited under subsection (b) and serves only to authenticate and provide context for the fact evidence being offered. (e) No Impact on Other Computer or Machine Generated Outputs as Fact Evidence. Any other types of outputs from computers or other mechanical and/or electrical devices may be admitted as ordinary fact evidence when relevant and properly authenticated under Rules 401, 403, and 901. The party offering such evidence must establish its authenticity and relevance, and the court must ensure that the probative value is not substantially outweighed by any risk of unfair prejudice, confusion, or misleading the jury. III. Problems with Current Proposed Rule 707 The Advisory Committee should reject the current proposed Rule 707 in favor of one that prohibits the use of AI-generated outputs as expert testimony because the current rule: (1) creates a false equivalence between AI-generated outputs and human expert testimony under Rule 702; and (2) ignores the “black box” problem inherent in current AI systems. A. The False Equivalence Problem The central flaw of proposed Rule 707 is its implicit suggestion that AI-generated outputs can serve as a direct substitute for human expert testimony under Rule 702. This approach fundamentally misunderstands both the nature of modern AI systems and the role of expert witnesses in adversarial proceedings. Rule 702 permits expert testimony when specialized knowledge will assist the trier of fact but requires that such testimony rest on a sufficient factual basis, be the product of reliable principles and methods, and reflect a reliable application of those principles to the facts of the case. Critically, human experts can be cross-examined about their methodology, their reasoning process, and the basis for their conclusions. They can explain gaps J. Redgrave, B. Redgrave, and T. Redgrave Comment on Proposed Rule 707 February 16, 2026 Page 6 in their analysis, acknowledge limitations, and adjust their opinions when confronted with new information. 6 AI systems, by contrast, cannot perform any of these testimonial functions. They cannot take an oath, be cross-examined, or reliably explain their reasoning in any meaningful way. 7 More fundamentally, they are unable to meaningfully possess and reliably employ the specialized knowledge that Rule 702 contemplates. Core to contemporary AI systems are models that are trained on observed data to approximate the unknown probability distribution representing the underlying data generating processes. AI systems use these internal models to calculate the likelihood/probability of each possible outcome based on the given input and then use these likelihoods/probabilities for tasks such as prediction or generation. Because of the probabilistic nature of this construction, AI systems do not know specific facts but instead have probabilistically informed “beliefs” or positions. 8 Exacerbating this epistemic failure, even when the output of AI systems may be correct or initially mirror that of a human expert, minor semantic or contextual changes 9 can often cause them to adopt an incorrect position, thereby demonstrating a lack of (robust) informed deterministic reasoning. 10 Perhaps most problematically, it has been shown that (language) models are guaranteed to produce hallucinations, even within an ideal scenario. 11 These unavoidable flaws significantly hinder the ability of AI systems to meet the reliability requirements of Rule 702. 6 Recent research indicates AI’s shortcomings in these areas may be one of the primary causes of AI hallucinations See, e.g., Why Language Models Hallucinate (https://doi.org/10.48550/arXiv.2509.04664). 7 Efforts to use AI models to explain themselves through “chain of thought” or so called “reasoning” models has been shown to be wholly insufficient, with models often providing misleading or incorrect explanations of their own behavior and even going as far as to obfuscate their reasoning when penalized. See, e.g., Reasoning Models Don't Always Say What They Think (https://doi.org/10.48550/arXiv.2505.05410); Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation (https://doi.org/10.48550/arXiv.2503.11926). For additional criticism of these terms broadly see Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces! (https://doi.org/10.48550/arXiv.2504.09762). 8 A troubling corollary of this result is that AI systems are incapable of categorically distinguishing between facts and belief. See, e.g., Language Models Cannot Reliably Distinguish Belief From Knowledge and Fact (https://doi.org/10.1038/s42256-025-01113-8). 9 See, e.g., Illusions of Confidence? Diagnosing LLM Truthfulness via Neighborhood Consistency (https://doi.org/10.48550/arXiv.2601.05905) and Representational and Behavioral Stability of Truth in Large Language Models (https://doi.org/10.48550/arXiv.2511.19166). 10 Ironically, indicating that the user prompting the AI system is an expert can exacerbate this issue. See, e.g., Epistemic Fragility in Large Language Models: Prompt Framing Systematically Modulates Misinformation Correction (https://doi.org/10.48550/arXiv.2511.22746). Adding insult to injury, recent work indicates that even when models encode a correct response, they may generate a false one (even without perturbing the input or context). See, e.g., LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations (https://doi.org/10.48550/arXiv.2410.02707). 11 See, e.g., Calibrated Language Models Must Hallucinate (https://doi.org/10.48550/arXiv.2311.14648) and On the Limits of Language Generation: Trade-Offs Between Hallucination and Mode Collapse (https://doi.org/10.48550/arXiv.2411.09642). The trade-off between hallucination and expressiveness is particularly relevant when considering specialized knowledge. J. Redgrave, B. Redgrave, and T. Redgrave Comment on Proposed Rule 707 February 16, 2026 Page 7 The Advisory Committee itself acknowledged these concerns in its explanatory materials, noting worries about “analytical error or incompleteness; inaccuracy or bias built into the underlying data or formulas; and lack of interpretability of the machine’s process.” Advisory Committee on Rules of Practice and Procedure, Agenda Book at 75 (June 10, 2025) (listing concerns with machine-generated evidence). The Advisory Committee further recognized that “[t]he hearsay rule is likely to be inapplicable because . . . a machine cannot be cross-examined.” Id. at 58. Yet despite acknowledging these fundamental problems, the proposed rule does not address them adequately. 12 B. The Black Box Dilemma and Interpretability Perhaps more problematic is the inherent opacity of modern AI systems—what researchers call the “black box” problem. When a neural network with billions or trillions of parameters generates an output, even its creators typically cannot trace the specific pathway of reasoning that produced that result. 13 While efforts to advance AI interpretability is a increasingly active area of research, and multiple paradigms for understanding models through various lenses have been developed, the field remains in its infancy; researchers can sometimes identify high-level patterns or features that influence AI behavior but cannot reliably explain why a particular input produced a particular output in any given instance. 14 This opacity is fundamentally incompatible with the reliability requirements of Rule 702. Rule 702 and its supporting case law require courts to evaluate whether an expert’s methodology is scientifically valid and properly applied. How can a court perform this gatekeeping function, or opposing counsel effectively cross-examine an AI output, when no one—not the offering party, not the system’s designers, not leading researchers in the field—can reliably explain the inferential steps that produced it? The answer is that they cannot. Proposed Rule 707’s failure to grapple with this reality renders it inadequate as a framework for admissibility and should be rejected. IV. Reasoning Behind the Proposed Alternative Approach In addition to adopting a rule that prohibits the use of AI outputs as expert testimony, the Advisory Committee should adopt a rule that frames AI as a tool rather than a witness while allowing AI-generated content to be admissible as fact evidence when proper foundation is established under the existing rules. 12 See Submission of James Beck for an expanded views of the issues surrounding the potential admission of AIgenerated “opinions” without supporting human testimony. (https://www.regulations.gov/comment/USC-RULESEV-2025-0034-0037); see also Submission of American Civil Liberties Union (https://www.regulations.gov/comment/USC-RULES-EV-2025-0034-0028). 13 See supra note 7. 14 For a brief overview of different approaches to AI interpretability as well as a survey of existing works exploring the mechanistic interpretability of AI models, see Mechanistic Interpretability for AI Safety -- A Review (https://doi.org/10.48550/arXiv.2404.14082). J. Redgrave, B. Redgrave, and T. Redgrave Comment on Proposed Rule 707 February 16, 2026 Page 8 A. The Appropriate Framework: AI as Tool, Not Witness A sounder approach to current proposed Rule 707 would recognize AI for what it is: a potentially useful tool that experts may rely upon, but not a substitute for human expertise. Under Rule 703, experts may base their opinions on facts or data that are not themselves admissible, provided those facts or data are of a type reasonably relied upon by experts in the field. If an epidemiologist uses AI to analyze patterns in disease transmission data, or a financial expert employs machine learning to detect fraud patterns, these are legitimate uses of technology as an analytical aid. But the expert—not the AI—must render the opinion and stand behind it. Under this approach, for such reliance to be appropriate, the expert must satisfy the Rule 702(a)–(d) factors regarding their use of AI. Specifically, the expert must demonstrate that: (a) their use of the AI tool will help the trier of fact; (b) they possess the qualifications to evaluate and interpret the AI tool’s outputs; (c) their testimony rests on a sufficient basis, including validation of the AI tool’s reliability for the specific task; and (d) they have reliably applied the AI methodology to the case facts. This framework preserves judicial gatekeeping while allowing experts to leverage technological tools where appropriate. B. AI-Generated Content as Fact Evidence: A Workable Alternative Under Existing Rules Separately, certain AI-generated content can and should be admissible as fact evidence when proper foundation is established. No new rule is required to accomplish this, however: if a litigant seeks to introduce AI-generated communications, documents, or other outputs not as expert opinion but as facts relevant to the case, the traditional evidence rules provide adequate safeguards. Under Rules 401 and 403, such evidence must be relevant, and its probative value must not be substantially outweighed by unfair prejudice or confusion. Authentication requirements under Rule 901 ensure that the evidence is what its proponent claims. Business records and other hearsay exceptions may apply where appropriate foundations are laid. For example, if the issue is whether a company deployed an AI chatbot that made false representations to consumers, the chatbot's outputs are admissible as fact evidence of what representations were made—not because the AI is testifying, but because those outputs are relevant facts in the case. Similarly, AI-generated content may be relevant to show a party’s reliance on AI advice, or the state of AI technology at a relevant time. In these scenarios, the AI-generated output is simply evidence like any other document or business record, subject to the ordinary rules of authentication, relevance, hearsay, and prejudice. J. Redgrave, B. Redgrave, and T. Redgrave Comment on Proposed Rule 707 February 16, 2026 Page 9 C. Proposed Alternative Rule 707 Does Not Require Separate Disclosure or Coordination with Civil Rules During the public comment process, including testimony, there was a spirited debate regarding the need for additional pre-trial disclosures of AI-generated evidence that potentially would be offered as “freestanding” expert opinion. That concern is real under the rubric of the current proposed Rule 707 and should not be resolved solely by reliance on proper case management. Under our proposed alternative, however, there is no need for disclosure because the AI-generated content, standing alone, can never be offered directly or indirectly as an “expert opinion” that would need a separate disclosure regime. 15 Of course, to the extent that a disclosed human expert is considering or relying on AI-generated outputs to form or support the opinion being proffered, then we expect that the rigor of existing Rule 702 will be applied to such proposed use. V. Thoughts on Other Rules Addressing AI-Generated Outputs Finally, we concur with submissions that recommend a go-forward approach to broader rulemaking to account for the current and likely future impact of AI-generated outputs on civil and criminal matters. 16 The issues surrounding “deep fakes” alone warrant rulemaking, but so do other considerations such as potential pre-trial disclosures regarding AI-generated content use in connection with civil and criminal litigation and other potential refinements to evidence rules regarding Hearsay (Article 8), Authentication and Identification (Article 9), and Contents of Writings, Recordings, and Photographs (Article 10). Concurrent with a republication of a revised proposed Rule 707 we recommend that the Committee on Rules of Practice and Procedure establish a working group among the five advisory committees that can focus on these other areas as well as obtain more feedback from judges, lawyers, and litigants with respect to whether AIgenerated outputs that are not within the scope of any enacted Rule 707 are being excluded or admitted into evidence under other rules that creates inconsistent, unfair, or unjust results. Finally, to the extent the Advisory Committee intends to publish a proposed rule on “deep fakes,” we recommend publishing that rule together with the republication of proposed Rule 707—as well as any other proposed civil and criminal rules that address similar subject matter—and posit that packaging the rules in this coordinated manner will elicit public feedback in a way that is more 15 The NYC Bar Association takes a different view, recommending that no evidence rule be adopted without corresponding revisions to Civil Rule 26 and Criminal Rule 16 to incorporate AI-specific discovery mechanisms. See Submission of the Federal Courts Committee of the New York City Bar Association at 3–4 (https://www.regulations.gov/comment/USC-RULES-EV-2025-0034-0046). While that concern is well-taken under the current proposed Rule 707—which would permit freestanding AI outputs as quasi-expert testimony without any expert disclosure obligation—our proposed alternative substantially eliminates the gap by requiring that AI outputs offered as expert-level analysis always be accompanied by a disclosed human expert subject to Rule 26(a)(2). 16 See, e.g., Submission from Hon. Paul Grimm (ret.) and Prof. Maura Grossman (https://www.regulations.gov/comment/USC-RULES-EV-2025-0034-0021); Submission of Hon. John Facciola (ret.)(https://www.regulations.gov/comment/USC-RULES-EV-2025-0034-0025). J. Redgrave, B. Redgrave, and T. Redgrave Comment on Proposed Rule 707 February 16, 2026 Page 10 respectful of both the Advisory Committees’ views on avoiding too frequent rule changes, and the courts and parties that will need to implement any newly adopted rules. CONCLUSION We respectfully submit that current Proposed Rule 707 fails to provide a proper framework in practice because it conflates three distinct categories of AI-generated evidence and creates an inappropriate pathway for AI outputs to serve as expert testimony, either directly or indirectly. The rule should be revised to clearly prohibit the introduction of AI-generated outputs as a substitute for human expert witnesses under Rule 702. At the same time, it should recognize that experts may appropriately rely on AI tools in forming their opinions, provided they can establish the reliability of that reliance under existing Rule 702 factors. Finally, the rule should acknowledge that AI-generated content may be admissible as ordinary fact evidence when relevant and properly authenticated under Rules 401, 403, and 901. This tripartite framework respects both the potential utility of AI technology and the fundamental requirements of reliability, transparency, and adversarial testing that underpin our evidence system. Until such revisions are made, Rule 707 in its current form should be rejected as inconsistent with the scientific and legal principles it purports to serve. Thank you again for the opportunity to submit our comments and suggestions on the proposed rule. /s/ Jonathan Redgrave Jonathan Redgrave /s/ Benjamin Redgrave Benjamin Redgrave /s/ Timothy Redgrave Timothy Redgrave An official website of the United States Government. < Back to Document Comments (/document/USC-RULES-EV-2025-0034-0001/comment) Share PUBLIC SUBMISSION Comment from Waters, Deborah Posted by the United States Courts on Feb 17, 2026 Docket (/docket/USC-RULES-EV-2025-0034) / Document (USC-RULES-EV-2025-0034-0001) (/document/USC-RULES-EV-2025-0034-0001) / Comment Comment See attached file(s) [ Attachments 1 2026.02.16 FRE 707 Comment Give Feedback Download (https://downloads.regulations.gov/USC-RULES-EV-2025-0034-0062/attachment_1.pdf) Comment ID USC-RULES-EV-2025-0034-0062 Tracking Number mlp-rslc-5r2o Comment Details Received Date Submitter Info WATERS LAW FIRM P.C. Town Point Center Building, Suite 600 150 Boush Street Norfolk, Virginia 23510 Deborah C. Waters Attorney at Law Proctor in Admiralty February 16, 2026 Committee on Rules of Practice and Procedure Administrative Office of United States Courts One Columbus Circle, NE Washington, DC. 20544 Re: Federal Rule of Evidence Rule 707 Comments on Proposed Amendment Dear Honorable Committee Members: The purpose of this letter is to comment briefly on proposed amendment to Federal Rule of Evidence 707. I am an attorney in Virginia in private practice who for the past 38 years has represented plaintiffs exclusively. Most of my cases now focus on admiralty and maritime law; thus, many of the cases are filed in various federal courts in districts across the country. In the past, I represented plaintiffs for diverse types of personal injury claims and for discrimination claims when the discrimination affected large groups of people. My practice also included participation in the BP Oil Spill Multi-District Litigation. Accordingly, I have experience with admissibility of evidence under the 700-series of the Federal Rules of Evidence. I have read the text of the proposed amendment to Rule 707, the comments posted about the proposed amendment, and the hearing testimony. The main reason I feel the need to comment on proposed amendments to Rule 707 is because I found it overly broad, ambiguous and confusing. Although it is difficult to fashion clear, concise rules of court, the Rules of Evidence should be clear enough to be workable. I believe the proposed amendment to Rule 707 needs more work to reach that level. Please reconsider the amendments in view of the excellent observations and suggestions made in comments filed with the Committee and by witnesses who testified in person at the hearings held about Rule 707. I do not believe the amendment should be adopted as written. The Committee would be wise to include in an amendment to Rule 707 more definitional clarity. For example, definitions of “machine generated evidence” and of “basic scientific instruments” would provide more clarity to language that is currently vague and ambiguous. At this point it seems overly broad. Litigants have long employed as evidence machine generated data maintained in the ordinary course of business and should be allowed to continue that practice. The test for admissibility is clear, and courts and litigants have long relied on the certainty generated by the rules. My concern is that the new language will make previously admissible evidence inadmissible. _____________________________________________________________________________________ Telephone: 757.446.1434 dwaters@waterslawva.com Facsimile: 757.446.1438 Committee on Rules of Practice and Procedure February 16, 2026 Page 2 Fortunately, or unfortunately, the technological advances we are experiencing in society today – the age of Artificial Intelligence - created a need for the Committee to consider amending the Rules of Evidence. As the Committee pointed out in its comments to the Rule, Artificial Intelligence has created a new category of information, because the information now can rise to the level of opinion in the past provided only by human beings. The Committee is wise to consider computer generated opinions in context of the Rules of Evidence to determine whether the existing rules are sufficient. The proposed amendment has been the subject of extensive comment in the record. It is interesting to note that nearly every commentator from every perspective urges the Committee to withhold adoption of the proposal and give it further consideration. I agree. I will not write extensive comments because of redundancy, but support those in the record made by Steve Herman and Tad Thomas. Accordingly, I respectfully request the Committee table the amendment to Rule 707 and continue to work with the public to fashion one or more workable rules or have further discussion about whether an amendment is needed. Respectfully, Deborah C. Waters DCW/ _____________________________________________________________________________________ Telephone: 757.446.1434 Email: dwaters@waterslawva.com Facsimile: 757.446.1438 An official website of the United States Government. < Back to Document Comments (/document/USC-RULES-EV-2025-0034-0001/comment) Share PUBLIC SUBMISSION Comment from National Association of Criminal Defense Lawyers Posted by the United States Courts on Feb 17, 2026 Docket (/docket/USC-RULES-EV-2025-0034) / Document (USC-RULES-EV-2025-0034-0001) (/document/USC-RULES-EV-2025-0034-0001) / Comment Comment Attached you will find the comments of the Nat'l Ass'n of Criminal Defense Lawyers on the proposed new FRE 707. Attachments 1 NACDL Comments on Proposed Federal Rule of Evidence 707 Download (https://downloads.regulations.gov/USC-RULES-EV-2025-0034-0063/attachment_1.pdf) Give Feedback r Comment ID USC-RULES-EV-2025-0034-0063 Tracking Number mlp-t9sb-thqk The Honorable Jesse M. Furman, Committee Chair Professor Daniel Capra, Reporter Members of the Advisory Committee on the Evidence Rules February 16, 2026 Submitted online Re: Proposed Federal Rule of Evidence 707 Dear Judge Furman, Professor Capra, and Members of the Advisory Committee on Evidence Rules: The National Association of Criminal Defense Lawyers (NACDL) is pleased to submit our comments on the proposed new Federal Rule of Evidence 707. Our organization has a nationwide membership of thousands of direct members, and up to 40,000 other members through state affiliates. NACDL’s members include private criminal defense attorneys, public defenders, military defense counsel, law professors, and judges. The majority of our members are trial lawyers who are regularly litigating in both state and federal courts around the country. Thus, NACDL is in a unique position of having input from the ongoing experiences of lawyers with substantial experience with the challenges posed by artificial intelligence and other machine-generated evidence. Currently, NACDL’s Fourth Amendment Center is litigating technologies that include, but are not limited to, automated license plate readers, geo-fence warrants, mass cell-site location information, facial recognition technology, ShotSpotter, and forensic genetic genealogy. I. Introduction Given our organization’s experience with these issues, we are keenly aware that “machine-generated evidence” and how courts should address it are important to both the general public and the law. While NACDL is pleased that the Advisory Committee on Evidence Rules (“Advisory Committee”) is taking steps to address this emerging area of the law, we have concerns about the proposed FRE 707 (P.F.R.E 707) as currently written and encourage the Advisory Committee to pause before adopting it. 1660 L Street NW, 12th Floor, Washington, DC 20036 / () lacebook.com/NACDL O twitter.com/NACDL 202-872-8600 / ® instagram.com/NACDL assist@nacdl.org NACDL.org NACDL.org/ Foundation Given the quick pace of technological change, a pause would give the Advisory Committee ample time to consider the ramifications of this proposed rule, it's impact in our courtrooms around the country and on determinations that affect life and liberty. If the Advisory Committee is inclined to move forward with the proposed rule at this time, NACDL would like to offer suggestions to clarify the rule and ensure that it accurately reflects the Advisory Committee's intent. II. The Advisory Committee should pause before moving forward with this proposal at this time. History shows us that technology evolves faster than the law can keep up. Since the Advisory Committee began considering an amendment to address machinegenerated evidence, it has been keenly aware that this technology has been evolving rapidly. Private companies currently developing artificial intelligence (AI) systems claim uncertainty about AI consciousness1, while scholars caution that anthropomorphizing these technologies “serves to obscure the actions and accountability of people building and using the systems”.2 We simply do not yet know how these technologies will be presented in court. As other commentators have noted, the rule, as currently proposed, raises serious issues about applicability and feasibility. Adopting this rule, as written, at this time, risks cementing outdated technological concepts in the Federal Rules of Evidence. Considering that the Advisory Committee disfavors amending the same rules multiple years in a row, adopting this proposal would leave federal practitioners with an outdated rule for several years.3 1 Frank Landymore, Anthropic CEO Says Company No Longer Sure Whether Claude Is Conscious, FUTURISM (Feb. 14, 2026), available at https://futurism.com/artificial-intelligence/anthropic-ceo-unsure-claude-conscious 2 Emily M. Bender & Nanna Inie, We Need to Talk About How We Talk About “AI”, TECH POLICY PRESS (Jan. 7, 2026), available at https://www.techpolicy.press/we-need-to-talk-about-how-we-talk-about-ai/ 3 Memorandum from Daniel J. Capra on Amendments to Rule 609(a)(1) to Advisory Comm. on Evidence Rules (Oct. 1, 2024), in ADVISORY COMM. ON EVIDENCE RULES NOV. 8, 2024 AGENDA BOOK at 320, available at, 202411_evidence_rules_committee_meeting_agenda_book_final_10-24.pdf 1660 L Street NW, 12th Floor, Washington, DC 20036 / (} focebook.com/NACDL O twitter.com/NACDL 202 -872-8600 / ® instagrom.com/NACDL assist@nacdl.org NACDL.org NACDL.org/Foundotion Additionally, states have historically adopted the Federal Rules of Evidence into their own evidence rules.4 Adopting proposed Rule 707 now would compound the problem and leave states with a rule that is difficult to modify as technology evolves. For these reasons, NACDL urges the Advisory Committee to act with caution before promulgating any amendment on machine-generated evidence. III. A. If the Advisory Committee elects to go forward with promulgating a rule, NACDL would like to raise the following areas of concern: This rule risks subsuming the Federal Rule of Evidence 702. The current text of P.F.R.E 707 assumes that evidence will be presented without an expert. It doesn’t require expert testimony but rather explicitly envisions and creates a situation that governs the admissibility of evidence without experts. Lawyers for Civil Justice has called this a “pathway for admission” in their comments to the Advisory Committee. The presumption underlying the rule is that evidence will be presented in court by non-expert witnesses. The rule provides that, in this scenario, a judge must determine whether the evidence presented meets the requirements of F.R.E. 702 without the assistance of an expert witness. The rule incentivizes the removal of experts from the trial process altogether. A major consideration for governmental agencies, including investigative agencies and prosecutors, is cost reduction. The logic of efficiency and convenience often undergirds the introduction of technology and tech solutions. The logic of costcutting is likely to apply to the expensive process of hiring expert witnesses. While the rule is not intended to encourage parties to opt for machine-generated evidence over expert witnesses, such a rule would inevitably result in encouraging cutting the cost of expert witnesses in favor of machine-generated evidence and lay witnesses. Experts are vital where technology is concerned, particularly in the criminal legal system, where technologies known to produce unreliable results often continue to be used. Where novel technologies are concerned, experts become even more important. For example, new complex technologies like AI applications are not created by a single person “possessing personal knowledge of all the facts that are needed to 4At this writing, Arizona, Kentucky, Louisiana, Michigan, and Ohio have adopted their own rules on expert testimony to match the amendment to Rule 702, which was made effective in December 1, 2023. 1660 L Street NW, 12th Floor, Washington, DC 20036 / (} focebook.com/NACDL O twitter.com/NACDL 202 -872-8600 / ® instagrom.com/NACDL assist@nacdl.org NACDL.org NACDL.org/Foundotion demonstrate that the technology and its output are what its proponent claims them to be. Data scientists may be required to describe the data used to train the AI system. Developers may be required to explain the features and weights that were chosen for the machine-learning algorithm.” (Grimm, Grossman, and Cormack) 5. Witnesses who operate these technologies, or lay witnesses who present the result and output in court, are likely to have limited knowledge as to how they are created and the testing that measures their reliability. When such witnesses present evidence, the burden is likely to fall on defense counsel to challenge the reliability of the technology. This extra burden disproportionately affects under-resourced individuals and places a heavy burden on public defenders, who regularly face budgetary crises and may not have the resources to retain experts for every criminal case. B. Instead of reversion to F.R.E. 702, proposed F.R.E. 707 should include the factors outlined in F.R.E 702(a)-(d) with modifications more appropriate to evidence instead of human expert analysis. Proposed F.R.E. 707, which contemplates the types of evidence produced by some kind of scientific, machine, or computing process, would better be served by modifying the factors contained in F.R.E. 702 to properly apply to any such evidence admitted through either an expert or a lay person. While there is still danger inherent in creating a rule that allows evidence to be admitted which should require an expert 6, if this rule change does go forward, proposed F.R.E. 707 should not simply refer back to the factors of F.R.E. 702, as 702 applies to evaluation of a human expert – not a computational output. As such, Proposed F.R.E. 707 should spell out with specificity the type of evaluation courts should apply when acting as the gatekeepers of the admission of such evidence. 5 See Paul W. Grimm,et al., Artificial Intelligence as Evidbattery-operated digital thermometers vary widely in complexityence, 19 NW. J. TECH. & INTELL. PROP. 9 (2021), available at, https://scholarlycommons.law.northwestern.edu/njtip/vbattery-operated digital thermometers vary widely in complexitybattery-operatedol19/iss1/2) 6 Committee Note acknowledges that this rule does not intend to discourage expert testimony, nor does it anticipate that those intending to introduce such evidence without an expert will be able to overcome the hurdle of the analysis required by this rule. With this being the case, it would seethe m that such a rule is indeed unnecessary and that any such evidence should require an expert or be otherwise entered through stipulation. See Committee Note, at 61-73, 1660 L Street NW, 12th Floor, Washington, DC 20036 / (} focebook.com/NACDL O twitter.com/NACDL 202 -872-8600 / ® instagrom.com/NACDL assist@nacdl.org NACDL.org NACDL.org/Foundotion F.R.E. 702 provides a good foundation for the types of inquiries the court must make but runs into hurdles when applied to a computational output. The Committee Note, at 48-55, states the following: The rule applies when machine-generated evidence is entered directly, but also when it is accompanied by lay testimony. For example, the technician who enters a question and prints out the answer might have no expertise on the validity of the output. Rule 707 would require the proponent to make the same kind of showing of reliability as would be required when an expert testifies on the basis of machinegenerated information. The Advisory Committee is indeed addressing the right concerns about the attempts to admit computer-generated outputs through lay witnesses; however, the proposed evaluation of this evidence type through F.R.E. 702(a)-(d) by the courts requires refinement. Taking the example of the lay technician who does not have the requisite expertise in the validity of the output, the rule does not explain how the technician, or the party seeking to admit the evidence, should apply the factors in 702(a)-(d) to make a showing of reliability as it relates to the generated output they seek to admit. The factors in F.R.E. 702(a)-(d) state the following: A witness who is qualified as an expert by knowledge, skill, experience, training, or education may testify in the form of an opinion or otherwise 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. Those factors assume that there is a person who intends to admit the evidence and who has specialized knowledge to answer the questions posed in the evaluation 1660 L Street NW, 12th Floor, Washington, DC 20036 / (} focebook.com/NACDL O twitter.com/NACDL 202 -872-8600 / ® instagrom.com/NACDL assist@nacdl.org NACDL.org NACDL.org/Foundotion of the evidence. Further, what kinds of information would the courts need to see to feel comfortable that the computer-generated outputs are not only reliable but also could be properly understood by the fact-finders if presented by such a lay technician? The Committee Note provides some examples but does not entirely contemplate the problem of how such information would be produced and conveyed. Who would produce and explain evidence relating to the training data sets used by the algorithmic program in question? What kinds of reliability data would be provided and by whom? Ultimately, it seems that it would be impossible for any lay witness to meet the standards anticipated by this rule and such evidence would require some form of expert testimony. C. Clarity of terms is necessary because the term “machine-generated” is so broad that it would apply to most evidence introduced in courts today. As other commenters have noted, the term “machine-generated” is so broad as to apply to a huge array of evidence being introduced in the criminal legal system. Because of the broadness of the term, P.F.R.E. 707 risks becoming the default for admission of evidence that is generated by machines, is the output of machines, or even has machines involved in the process at any step. Any rule attempting to define evidence that is the result of a computational or scientific process should use industry-standard terminology that is widely accepted and understood. 1. The addition of language intending to exclude “the output of simple scientific instruments” as a means of clarity and efficiency, instead creates further confusion about the application of this rule. The text of the proposed rule includes a caveat relating to the “output of simple scientific instruments” in an attempt to define the types of evidence the rule seeks to apply to and avoid unnecessary litigation in the courts. While the Advisory Committee note provides examples of what a “simple scientific instrument” might constitute7 [Text truncated.] --- Source: Frix Law Library, https://www.frixlaw.com/law-library/documents/federal_court_rules%3Aus-national%3Abc1a412bc6d9320d. Public record. Not legal advice.