Evidence Excluded In Murder Case for Failure to Disclose AI Use

The future of AI use in criminal investigations will mean more transparency.

As I have repeatedly written in posts here, AI is a tool, it is not a product. As such, especially in criminal cases, knowing where it is used, what its limits are and being able to explain that to the court is critical for some cases. In the case highlighted in this post, it could mean the difference between life in prison and a dismissal of the case.

The case is the State of Ohio v. QEYEON TOLBERT, CR-24-689572-A. The case is currently pending in the Cuyahoga County Court of Common Pleas. You can review the public docket here.

Tolbert was indicted for murder. His initial attorney was permitted to withdraw and new counsel, Brian Fallon, was appointed. Mr. Fallon’s first filed motion was a motion to suppress the firearm and other evidence law enforcement discovered during a search of Tolbert’s residence. The discovered evidence is what connected Tolbert to the murder according to prosecutors.

Mr. Fallon’s motion to suppress that evidence contained the following:

On February 20th, 2024, Det. Legg #2198 of the Cleveland Police Department obtained a search warrant for 403 E. 152nd St., Cleveland, Ohio, apartment #1, the residence of the defendant in the present case. The affidavit for the search warrant contained 24 numbered paragraphs outlining the "facts” upon which the affiant based his beliefs. See Ex. #A attached. The defendant is referred to as the “unidentified male” from paragraphs 10-19 in the affidavit. Then in paragraph 20 it is sworn as follows:

“Affiant avers that utilizing the Fusion center they received an identification of the, as of yet, unidentified male suspect, based on the recovered surveillance video, and it was learned this male was currently paroled to the address of 403 E. 152nd Street, Apartment #1.”

This is the entire basis of how the police identified the defendant and were able to obtain a search warrant for his residence.

The above quoted paragraph 20 is entirely misleading if not an outright falsehood.

The reason for Mr. Fallon’s assertion that this provision of the search warrant was false is that the reference to the “Fusion Center” obscured the fact that it was using facial recognition software from Clearview AI to identify Tolbert and provide the key information substantiating the search of his residence.

Summary of Clearview AI

Clearview AI is a technology company that provides facial recognition software primarily for law enforcement and government agencies. Its core product is a powerful facial recognition tool that allows users to identify individuals by comparing photos to a database of publicly available images scraped from the internet, including social media platforms, news websites and other online sources.

Who Cares If Clearview AI’s Facial Recognition Software Was Used?

Mr. Fallon read published Clearview AI materials on its website where it contained the following information about the accuracy and reliability of its AI powered facial recognition software.

“DISCLAIMER: Facial recognition search results are to be treated as investigative leads and should not be solely relied upon for making an arrest. Investigators are required to conduct thorough investigations, independently verify identifications, and adhere to applicable laws and agency policies regarding facial recognition search results. The information in this document is for internal use only and should not be shared outside the user’s agency. These search results are not intended or permitted to be used as admissible evidence in a court of law or any court filing,”

“As with any search engine, search results established through Clearview and its related systems and technologies are indicative and should not be considered definitive. Clearview makes no guarantees as to the accuracy of its search identification software. Clearview’s facial recognition algorithm has been tested by the National Institute of Standards and Technology’s Facial Recognition Technology Evaluation program, and was found to be highly accurate, but its performance under real world conditions can differ. The quality of a submitted probe image, the lack ofonline images of a depicted individual in Clearview’s Database, and other factors can impact and potentially reduce the accuracy of the Clearview search results. A set of search results produced by the Clearview search engine may contain a mix of images of the person depicted in the probe image and images of other similar looking individuals. Clearview is neither designed, nor intended, to search the internet for artificially generated images of faces to be used as a sole source system for conclusively establishing or determining an individual's Identity, it is the responsibility of the Customer to corroborate any identifying information or other data discovered on third party sites using any Clearview system or included in Clearview search results by conducting additional research.”

Based upon this, Mr. Fallon’s motion contained the following assertion to the court:

“The affidavit states in paragraph 18 that six days after the murder, on February 20, 2024 the police watch video from the fusion center and see a male walking from the area of 403 E. 152nd Street and the affidavit claims the male has the same build, haor style clothing and walking characteristics as the defendant. This is the video on the 20th taken from a convenience store that the police claim the Fusion Center identified the defendant, not from the surveillance video from the night of the shooting. As we now know, the Fusion center cannot use Clearview ai facial recognition to identify anyone definitively.”

The search warrant argued that the affidavit claimed an identification as made by use of Clearview AI’s tool when the company making the tool disclaims any ability to make a reliable, positive identification.

The affidavit failed to disclose police had used this facial recognition tool and failed to disclose the lack of reliability of that tool.

Following a hearing on the Defendant’s motion, the court excluded all evidence obtained from Mr. Tolbert’s residence using that search warrant.

The decision by the trial court to exclude the evidence is now being appealed by the state.

The lawyers for Mr. Tolbert told the local media source Cleveland.com that “It really wasn't a positive identification” which made the affidavit to the search warrant misleading.

This case has potential impact on hundreds of other cases. For the past 2.5 years, the Ohio Bureau of Criminal Investigation, relied upon by the Ohio Attorney General’s Office, reports that it has used Clearview Al in nearly 200 investigations. Even that office has produced a disclaimer similar to Clearview AI’s stating that “Facial recognition is an investigative tool only and does not establish probable cause to obtain an arrest warrant without further investigation.”

Here are five effective practice tips for criminal defense lawyers representing clients whose cases involve the use of Clearview AI’s facial recognition technology during the investigation:

1. Challenge the Validity of the Identification

• Examine the Accuracy of the Technology: Investigate Clearview AI’s error rates, especially for racial, ethnic, and gender biases, which may lead to false identifications. Studies have shown that facial recognition technologies can disproportionately misidentify certain groups.

• Demand Discovery: Request all records related to the use of Clearview AI in the case, including:

• The specific image(s) used to query the database.

• The database query results and any matches provided.

• Internal policies or procedures used by law enforcement for confirming Clearview matches.

• Engage Experts: Retain a facial recognition expert to evaluate the reliability of the identification and whether it aligns with industry standards.

2. Raise Constitutional Challenges

• Fourth Amendment: Argue that the use of Clearview AI to conduct a search or initiate an investigation constitutes a warrantless search, potentially violating your client’s right to privacy.

• Investigate whether law enforcement relied solely on facial recognition to establish probable cause.

• Fourteenth Amendment: Highlight the potential for systemic bias in facial recognition technology to violate your client’s due process rights if the technology played a determinative role in the case.

3. Scrutinize the Evidence Chain

• Focus on Probable Cause: Examine whether Clearview AI’s output provided the sole or primary basis for searches, arrests, or further investigations. Argue that reliance on unreliable technology undermines the validity of subsequent warrants or evidence obtained.

• Challenge the Chain of Custody: Ensure that law enforcement properly documented the use of facial recognition, avoiding improper shortcuts or misrepresentations.

• Fruit of the Poisonous Tree Doctrine: If Clearview AI’s identification process was flawed or improperly used, argue for the exclusion of all downstream evidence as tainted.

4. Leverage Discovery and Suppression Motions

• File motions to obtain detailed information about:

• The agreement or contract between law enforcement and Clearview AI.

• Training materials provided to officers using the tool.

• Internal audits, error logs, or usage reports from the agency.

• Use this information to build motions to suppress evidence based on procedural flaws, constitutional violations, or lack of reliability in the underlying technology.

5. Educate the Court and Jury

• Simplify the Technology: Provide the court and jury with clear, understandable explanations of how facial recognition works, its limitations, and its known errors. Use analogies and visual aids, if possible.

• Focus on Bias and Reliability: Present studies and expert testimony highlighting issues with Clearview AI, including:

• False positives and potential for misidentification.

• The lack of transparency in Clearview AI’s proprietary database and algorithms.

• Highlight Overreliance: Emphasize to the jury or judge that law enforcement’s reliance on unverified AI technology may have led to improper or hasty conclusions.