AI-Generated Evidence in Michigan Courts (2026)
Generative AI now produces voice clones, deepfake video, AI-summarized medical records, and AI-reconstructed accident scenes that can look indistinguishable from real evidence. Michigan trial courts are addressing these issues under existing authentication standards in MRE 901 and the reliability framework of MRE 702. A proposed amendment to the Federal Rules of Evidence would add a dedicated AI provision, but it is not yet in effect, and Michigan has not adopted parallel language.
Trending Hooks Driving the 2026 Caseload
The viral moments are familiar: a clip of a politician saying something they never said, a voice-clone scam, a crash reconstruction that turns out to be AI-generated rather than from dash-cam footage. The legal questions are less famous but more consequential. When the same generation tools are used to manufacture or alter evidence for litigation, courts must decide what authentication looks like.
Three recurring evidence problems are surfacing in Michigan trial practice in 2026: AI-summarized medical records offered in support of PIP claims; AI-cleaned dash-cam footage offered to enhance vehicle identification; and AI-generated images used to illustrate accident reconstructions. None of these are inherently inadmissible. All require a level of authentication and reliability foundation that the proponent often cannot meet without testimony explaining the tool.
MRE 901: Authentication Is the First Gate
Michigan Rule of Evidence 901 requires the proponent of an item of evidence to produce sufficient evidence to support a finding that the item is what the proponent claims it is. The rule is non-exclusive; it lists examples including testimony of a witness with knowledge, distinctive characteristics, and process-or-system evidence.
For AI-generated or AI-modified evidence, the realistic authentication path is process-or-system testimony from someone who can describe what the original input was, what the tool did, whether the original was preserved, and why the output reliably represents the underlying input. A treating physician who used an AI tool to summarize records can usually authenticate the original records and her own review; she cannot authenticate the AI summary as accurate without comparing it to the underlying records.
The controlling rule: MRE 901(a) requires authentication as a condition precedent to admissibility, satisfied by evidence sufficient to support a finding that the item is what the proponent claims. MRE 901(b)(9) specifically allows authentication by evidence describing a process or system and showing that it produces an accurate result. That subsection is the primary entry point for AI-generated evidence in Michigan trial practice.
Proposed Federal Rule 707 and What It Signals for Michigan
The Judicial Conference’s Advisory Committee on Evidence Rules has proposed a new Federal Rule of Evidence 707 addressing machine-generated evidence. As drafted, 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 it only if it satisfies the requirements of Rule 702(a)–(d). The rule does not apply to the output of simple scientific instruments.
The proposed rule was released for public comment in August 2025, the comment period closed in February 2026, and — if it follows the standard rulemaking track — it would not take effect before December 1, 2027. It is not yet enacted law.
Michigan has not adopted a parallel state-level rule. Michigan courts can and do borrow from federal authority interpreting analogous rules, particularly on technical evidence questions, so the framework behind proposed Rule 707 is useful persuasive material now. Watch for the Michigan Supreme Court to consider a corresponding amendment in the coming rule-amendment cycles.
Daubert Reliability Where Expert Testimony Underpins the AI Evidence
When AI output is offered through an expert witness, MRE 702 and the Daubert reliability framework apply. The expert must explain the method, the rate of error, validation, and general acceptance in the relevant field. AI tools that are proprietary, frequently updated, and trained on undisclosed data sets often struggle to meet that standard.
The practical effect is that AI evidence often does better when offered as demonstrative rather than substantive — an AI-generated reconstruction illustrating an expert’s testimony rather than serving as the substance of the testimony itself. Michigan courts can allow demonstrative use with a limiting instruction, even when the same content would not be admissible as substantive proof under MRE 702 and 901.
Deepfake Objections and the Burden of Proof
The most fraught category is the alleged deepfake. An opponent who claims an offered video or audio is AI-generated raises an authentication challenge under MRE 901 that the proponent must address. The proponent does not have to prove the impossible, but must come forward with sufficient evidence to support a finding of genuineness.
Useful Michigan-applicable indicators include chain-of-custody from the recording device, metadata showing time and location consistent with other evidence, preservation of the original file, and corroboration from witnesses who experienced the recorded event. AI-detection software exists and is improving, but no current tool is reliable enough to be admissible as proof of authenticity standing alone. Courts treat detector output as one input into the totality-of-the-circumstances analysis under MRE 901(a).
Practical Posture for Michigan Trial Practice in 2026
Litigants offering AI-generated content should preserve the original file, disclose the enhanced or generated version separately, identify the tool and process used, and plan to put on a witness who can explain why the output is reliable. Litigants challenging AI-generated content should make the authentication objection at the earliest opportunity, request the original file and metadata, and demand the foundation in advance of trial. Both sides benefit from a pretrial stipulation, where possible, on how AI summaries or reconstructions will be used. Saving these fights for trial guarantees a complicated voir dire of the proffered AI tool in front of the jury.
RELATED KNOWLEDGE BASE ARTICLES
AI Chatbots Posing as Lawyers: Michigan UPL Risk · AI Insurance Claim Denials in Michigan No-Fault · Serious Impairment Threshold After McCormick
Frequently Asked Questions
Is an AI-summarized medical record admissible in a Michigan PIP case?
It can be used in limited ways, but the summary is not a substitute for the underlying records. The proponent should offer the original records as the substantive evidence and use any AI summary as a demonstrative aid. Authentication of the summary itself requires testimony explaining how the tool was used and a comparison with the underlying records.
Can I object to a deepfake video on authentication grounds?
Yes, under MRE 901. Once you raise a credible objection, the proponent must come forward with sufficient evidence of genuineness. Useful indicators include chain-of-custody, metadata, corroboration from witnesses, and the absence of artifacts characteristic of generation tools. A bare assertion without any indicator that the video is fabricated is unlikely to keep otherwise authenticated footage out.
Does proposed Federal Rule 707 apply in Michigan state court?
No. Proposed FRE 707 is a federal rule that is not yet in effect, and Michigan state courts apply the Michigan Rules of Evidence regardless. Federal authority on analogous questions is persuasive only. Whether Michigan adopts a corresponding state rule will likely depend on Michigan Supreme Court rule-amendment proceedings in the next several cycles.
Can AI-cleaned dash-cam footage be used?
Yes, if properly authenticated. The proponent should preserve and offer the original recording, explain how the enhancement was applied, and ideally call the engineer or expert who performed it. Courts are more likely to admit AI-enhanced video where the proponent can show the underlying recording is genuine and the enhancement clarified the original rather than adding content that was not present.
What if I receive an AI-generated demand letter?
Treat it as you would any demand: identify the actual claimant, evaluate the underlying claim on its merits, and respond. If the letter cites cases that do not exist or misstates Michigan law, those are facts to surface in your response. If a lawyer signed it, the lawyer is professionally responsible for the content regardless of how it was generated.
Have a question about this area of Michigan law?
Free, confidential review with Attorney Manny Chahal. Consultations available in English, Hindi (हिंदी), and Punjabi (ਪੰਜਾਬੀ).
Call 1-844-MCHAHAL for a free case evaluation