Is AI-Generated Evidence Admissible in Maryland Criminal Court?

Today, artificial intelligence (AI) is part of daily life. It is the power behind virtual assistants, is constantly seen on social media feeds, and even helps organize and streamline daily routines through smart home devices. It was certainly not going to be long before AI made its way into criminal proceedings. Now, a court in the state of Washington has made one of the first formal decisions in the country regarding the admissibility of AI-generated evidence. Below, our Maryland criminal defense lawyer explains the most relevant case law regarding the issue.

The Daubert and Frye standards

Before understanding the decision in the Washington case, it is important to understand the Daubert and Frye standards. These standards derive from two other notable cases.

The first is Daubert v. Merrell Dow Pharmaceuticals, Inc., which was heard by the U.S. Supreme Court in 1993. In this case, the Supreme Court outlined specific criteria for trial courts to use when determining if scientific knowledge, including novel technology, is admissible. The criteria outlined are as follows:

• Whether the theory or technique has been tested
• Whether the theory or technique has been published in a reputable publication, and if it has been peer-reviewed
• The known error rate for the technique or theory
• Whether standards exist to control the operation of the theory or technique
• If the theory or technique has gained widespread acceptance in the relevant scientific community

The Frye standard stems from the case Frye v. United States, which was heard and decided on by the United States Court of Appeals for the District of Columbia Circuit in 1923. This case established standards to determine if scientific knowledge stated by an expert in the relevant field was generally accepted within that field.

Some states today follow a hybrid model of using both the Daubert and Frye standards. In Washington, the Frye standard is used, while today, Maryland uses the Daubert standard in state and federal courts. The Supreme Court of Maryland officially abandoned the Frye standard in August of 2020, favoring the standard set in the Daubert case instead.

The Washington case

One of the first notable court decisions regarding whether AI-generated evidence is admissible in criminal courts was decided in the case Washington v. Puloka by the Superior Court of Washington for King County in March of 2024.

The case involved a shooting that resulted in a triple homicide. A witness recorded the incident on their smartphone and later streamed it on social media. The video was ten seconds long and, according to an expert witness for the defendant, was fuzzy and of low quality. The expert also testified that movements in the video were blurred. The expert used AI technology to improve the quality of the recording, and the defendant tried to admit the improved version of the video as evidence.

The expert in the case had been a professional filmmaker and videographer for over 20 years at the time. However, he had no formal training in forensic video analysis, and he did not have a background in the field.

The expert had used Topaz Labs AI to initially add definition and sharpness to the video. He then processed the video with Adobe Premiere Pro so the defendant could admit the highest quality of the video possible.

The expert was unable to state whether Topaz Labs AI is used by forensic video analysis professionals. He did not know if his peer group had tested the tool, published work regarding it, or evaluated its reliability for the purpose of video enhancement. He also could not state whether the program used algorithms or which large language model the AI tool used. The defendant argued the court should recognize the videography and filmography communities as relevant for purposes of the Frye standard.

The prosecution questioned their own witness, who was a Certified Forensic Video Analyst with credentials at the national and international levels. The prosecution’s expert witness criticized the defendant’s use of the AI tool to increase the pixel count of the source video. He also stated that the AI tool utilized an algorithm and enhancement features that have not been studied or approved by the professional forensic video analysis community.

Additionally, the state’s expert outlined how the AI tool did more than simply enhance the video. He stated that it altered the original video by removing data, eliminating blurred motions, and smoothing the edges of images. He also stated that the AI tool included additional information, which altered the color and shapes in the original video. Ultimately, the state’s expert testified that the AI tool made a proper and accepted video forensic analysis impossible.

Lastly, the expert for the state testified that the AI tool used was never tested, and so, the outcomes of those tests could not be published. As such, he stated that any findings from the video after AI was applied could not be evaluated by the expert witnesses for the defendant or the state, the court, or the forensic analysis community.

Ultimately, the court decided to exclude the defendant’s video that was enhanced by the AI tool, as well as the testimony of the defendant’s expert witness. When applying the Frye standard, the court found that the AI tool in question had not achieved general acceptance in the relevant scientific community. The court determined that the relevant scientific community in this case was the forensic video analysis community and not that of videographers and filmmakers.

The court also determined that allowing the admission of the enhanced video would further complicate the issues, confuse eyewitness testimony, and that the risk of unfair prejudice was greater than any relevance the enhanced video had on the case. The court determined that the best evidence for the jury to consider was the original video.

How the Daubert standard would apply in the case

While Washington courts do apply the Frye standard, the courts in Maryland exclusively apply the Daubert standard (along with other evidence rules governing authentication, relevance, and admissibility). Still, in this case, it is unlikely the outcome would have been different under a Daubert analysis because the defendant was unable to show that the tool had been tested and analyzed with results published in a reputable publication.

However, this is not to say that all AI-generated evidence is automatically inadmissible in criminal cases. It does outline the standards that must be met when defendants want to introduce AI-generated evidence in court and provides clear guidelines on admissibility. It also helps criminal defense lawyers navigate the intricacies of legal standards to present compelling expert testimony that is considered admissible.

Our criminal defense lawyer in Maryland can help you collect strong evidence

At Carey Law Office, our Maryland criminal defense lawyer can advise you of the laws and legal standards that will be applied in your case, and help you obtain the strong evidence that will help your case and shield you from the harshest outcomes. Call us today or contact us online to speak to our experienced attorney.