AI-Generated Evidence in Court: How Michigan Courts May Handle Deepfake and Synthetic Media

The development of artificial intelligence (AI) is growing rapidly, particularly in deepfake and synthetic media. And it's forcing courts to confront questions that just a few years ago seemed purely hypothetical. According to Act 11 of 2025 (the Protection from Intimate Deep Fakes Act), a “deepfake" refers to audio, video or digital content that has been manipulated or generated by AI. 

And the intention is to depict people saying or doing things they never said or did. Synthetic AI evidence includes computer-generated images, audio or simulations offered in court. In Michigan criminal cases, well-established rules of evidence and procedure govern the admission of evidence. 

And as deepfakes become more convincing and more common, Michigan trial judges and appellate courts will have to decide key factors. These factors include whether deepfakes can be admitted at all and what standards apply to determine authenticity. They'll also have to decide how juries should be instructed and how courts should handle deepfake challenges on appeal. 

Mark Linton is a precision-minded Michigan appellate advocate who has won numerous cases over the past two decades. He wrote this article to explain how Michigan courts may treat deepfake or synthetic AI evidence at the trial level and on appeal.

How Michigan Courts Evaluate AI-Generated Evidence in Court 

Michigan law states that all evidence must be both relevant and authenticated before it can be admitted at trial. This applies to a photograph, a video, or even an AI-generated video. 

Relevance Under MRE 401 and 402

Michigan Rule of Evidence (MRE) 401 gives clear guidance. It states that evidence is relevant if it has “any tendency to make the existence of any fact that is of consequence to the determination of the action more probable or less probable than it would be without the evidence.”

MRE 402 provides that relevant evidence is generally admissible unless excluded by law. A deepfake or synthetic video could be relevant if it tends to prove or disprove a fact at issue. For example, it may show whether or not a defendant was seen at a specific location. But relevance alone doesn't end the inquiry. 

Authentication Under MRE 901 and 902

Before admitting any evidence, Michigan courts require authentication MRE 901(a). This means a foundational showing that the item is what the proponent claims it to be. For traditional recordings, this often means witness testimony that the recording accurately reflects what it purports to show. 

For deepfake or synthetic evidence, authenticity is not simply whether the file can be traced to a source. Instead, it's whether it truly represents what it claims.  Let's say a prosecutor wants to introduce a video that allegedly shows a defendant committing a crime. In such a scenario, the court will need proof that the video is real and not manipulated. And the evidence could be metadata analysis or a forensic expert's testimony. 

There's no specific deepfake rule in the Michigan rules of evidence. But courts are already applying the existing authentication requirements to new technologies. However, Michigan's Act 11 of 2025 and subsequent updates have focused heavily on criminalizing the creation of deepfakes. This has, in turn, provided more statutory definitions that judges use when applying MRE 901 in criminal trials. 

Why Deepfake and Synthetic Media Create Authentication Problems” 

Authentication is typically established through testimony or technical verification. But in deepfakes, even expertly crafted videos can fool human observers and forensic tools. MRE 901(b) lists examples of authentication techniques relevant to deepfakes. Some of them include:

- “Testimony of a witness with knowledge” that the evidence is what it claims to be.  And this could be someone who viewed the original event.

- “Nonexpert opinion about handwriting, voice, or appearance.” This can be a witness recognizing a voice or face. But this can be undermined by a deepfake’s deceptive realism. 

- “Comparison by the trier of fact or expert witness” with an authenticated example. This is often key with digital media. 

- “Digital evidence authentication,” such as metadata or chain of custody, to show that a video has not been altered. 

A Michigan court will likely require stronger  proof before admitting a video or audio clip. This proof must show that the clip is not a deepfake. 

Authentication Challenges for AI-Generated Evidence in Court

Michigan courts have recognized the difficulty of authenticating sophisticated recordings in other contexts. In the People v. Berkey (1991) case, the Michigan Supreme Court held that evidence must be authenticated by showing it is what its proponent claims it to be. The Court emphasized that the threshold for authentication is not high. 

However, the proponent must still lay a foundation that the recording accurately shows the events or images claimed. And although the Berkey case involved traditional tape recordings, the principle applies to any visual media. In the Mitchell v. Kalamazoo (2017) case, the Court placed more emphasis. 

It insisted that the trial court must perform a “preliminary fact determination” before admitting evidence. This means the proponent must introduce sufficient proof that the item is genuine under MRE 901. This ensures that the evidence is at least reliable enough for a jury to consider.

Expert Testimony and AI-Generated Evidence in Court 

Michigan courts will likely require expert testimony under MRE 702 where deepfakes are at issue. This rule permits testimony from an expert under different conditions. These include when:

-The witness has specialized knowledge relevant to the case

-Their technical insight will help clarify issues for the fact-finder 

-The testimony rests on adequate foundation of facts or data

-The testimony is the product of reliable principles or methods 

-The expert has applied the methodology to the case facts in a reliable manner

Deepfake authentication almost always requires specialists. These are digital forensic analysts who can examine metadata, compression artifacts, inconsistencies in lighting or motion, and traces of manipulation detectable only by software. 

Without an expert to build this foundation, a judge may exclude a video or audio that seems like a deepfake. This is because the proponent cannot show it is what it claims to be. 

Additional Michigan Rules Affecting AI-Generated Evidence in Court

MRE 901 and trial court standards like the Trial Court Guidelines and Standards for Digital Imaging govern the admissibility of automated records. They also require safeguards for electronic evidence. And although not specific to deepfakes, they emphasize two things. 

The first is the need for a documented process showing how digital files were created, handled, and stored. The second is proof that the evidence has not been altered. Deepfakes often arise from artificial generation rather than discrete human creation. This means compliance with the defined expectations will be central. 

Constitutional and Due Process Concerns 

A deepfake may be authenticated and relevant. However, its admission may raise due process problems under the Michigan Constitution and the U.S. Constitution. If a prosecutor introduces a synthetic AI video that is misleading, false, or manipulative, the defense may argue that there are violations in the admission.

They may argue that it violates the Due Process Clauses of the Michigan Constitution (Const 1963, art. 1, § 17) and the U.S. Constitution (Fourteenth Amendment). These prohibit conviction based on unreliable evidence. They may also claim the evidence violates MRE 403 by posing too great a danger of unfair prejudice. This lets a judge exclude evidence if its value is clearly outweighed by a risk of unfair prejudice. 

A convincing but falsified deepfake could be so prejudicial that a judge would exclude it entirely. This is the same as graphic images or inflammatory evidence. MRE 403 gives judges discretion to exclude relevant evidence that risks unfair prejudice or misleading the jury. And given that deepfakes are designed to deceive, a court may find admission would mislead jurors and distort the fact-finding process. 

Jury Instructions and Safeguard

If digital evidence is admitted, courts may need to give special instructions acknowledging the limitations of AI-generated media. Michigan has no specific deepfake instructions yet. So, judges may adopt standard instructions on the witness credibility and evidence evaluation. This is in accordance with MCR 2.512(D) and the Model Civil Jury Instructions (M Civ JI) 97.13.

These instructions help address possible manipulation, the jury's role as a fact-finder, and the forensic expert’s limitations. This protects due process and helps jurors weigh digital evidence responsibly. 

How Michigan Courts Review Deepfake Issues on Appeal

A trial court can make an error related to deepfake evidence. The error might involve admitting an unauthenticated video or improperly instructing the jury. That issue can be raised on appeal to the Michigan Court of Appeals and potentially to the Michigan Supreme Court.

Appellate courts review different issues under different standards of review. Authentication and foundation decisions are typically reviewed for abuse of discretion. This means the appellate court will defer to the trial judge unless the decision was outside the range of reasonable outcomes. And the People v. Cameron (2011) case is a good example. 

Questions of due process or constitutional violations, on the other hand, receive De Novo review. This means the appellate court examines the issue fresh. To raise deepfake evidence issues on appeal, your defense counsel must preserve the objection at trial. MRE 103 states that a party must object and state the specific ground unless it is apparent from the context. A general objection isn't sufficient. 

Practical Effects of AI-Generated Evidence in Michigan Criminal Cases

Consider a case where the prosecution offers a video purportedly showing the defendant committing a crime. The defense will demand proof that the video is not a fake and may call its own expert. 

Both sides may offer digital forensic experts to support or challenge authenticity. The judge may exclude the video if its risk of misleading the jury outweighs its value. 

If admitted, the judge may instruct jurors to consider the possibility of manipulation. This is similar to instructions on eyewitness identification reliability. If the trial court errs, the defense can raise this as an abuse-of-discretion or constitutional error on appeal. 

Digital evidence against you may feel engineered or implausible, but the law gives you tools to defend yourself. And the right Michigan appeals attorney gives you leverage. Reach out to Mark Linton today to make sure the record reflects the truth, not technology's distortion of it.