Kimberlee Williams wrongly arrested by police using facial recognition
Kimberlee Williams wrongly arrested by police using facial recognition
Occurred: June 2021
Page published: May 2026
An Oklahoma resident was wrongfully arrested and incarcerated for six months after Maryland police departments relied on an unverified, external facial recognition match to obtain arrest warrants for bank fraud. The incident raised fears about police use of facial recognition to make arrests and led to calls for the reform of police policies and protocols.
Kimberlee Williams, a mother of seven working at a methadone clinic in Lawton, Oklahoma, was arrested while accompanying her daughter on a DoorDash delivery to a military base. When military police conducted a standard ID check at the base checkpoint, they discovered Williams had an outstanding arrest warrant in Maryland, detained her, and called local police, who arrested her.
The warrants stemmed from fraudulent over-the-counter bank withdrawals at several Maryland bank branches - an unknown individual had impersonated account holders and withdrawn thousands of dollars. A SunTrust bank investigator had shared the suspect's surveillance photo on CrimeDex, an online community of law enforcement and corporate investigators, and an unidentified respondent "suggested, using facial recognition software" that Williams was the suspect. Williams had never been to Maryland before.
Williams said she spent 23 days in an Oklahoma jail before an officer from Maryland arrived to transport her to a jail in Montgomery County, Maryland, where she was imprisoned for more than three months. Though Montgomery County prosecutors dropped their charges in October 2021, she still faced pending charges in two other Maryland counties, and was transferred to a jail in Prince George's County, where she spent another two months fighting charges there and in Anne Arundel County. Those cases were finally dismissed in December 2021, after she had spent a total of six months incarcerated.
The consequences didn't end with her release: the ordeal turned her and her family's life upside down, and she lost her job because of the arrest. She continues to worry she might unknowingly attract law enforcement attention again for something she had nothing to do with.
Several compounding failures allowed the error to escalate into a six-month imprisonment:
No corroborating investigation. None of the detectives, after receiving the bank's lead, appear to have done any further investigation for independent corroboration, relying instead on their own visual comparison.
Minimal information passed to police. In a memo to Montgomery County police, the bank investigator stated only that Williams was "identified... using facial recognition software"; to the other two counties, he omitted even that. It is unclear who actually ran the facial recognition search, what software was used, or what database was searched.
Concealment from the court. The detective assigned to the case did not disclose to the magistrate that the identification was made by facial recognition technology or that the scan had been run by an unknown website user, instead claiming the ID had been confirmed by visually comparing photos.
Known technical weakness. Testing has repeatedly shown facial recognition systems exhibit higher rates of false matches when used on people of colour, women, older people, and young people, an established flaw regulators and vendors have not resolved.
Structural incentive to over-trust the match: Once a name is attached to a face, confirmation bias sets in. Officers reportedly told Williams "You're telling me, but you're not telling me the truth," despite her insistence.
For Williams, the case meant lost income, months of incarceration far from her family, and lasting anxiety about being wrongly targeted again.
For society, it illustrates a systemic accountability gap: facial recognition results can enter the criminal justice system via private companies and anonymous online tips, with no chain of custody, no disclosed methodology, and no verification standard - yet can still result in an arrest warrant.
For policymakers, it strengthens the case for legally requiring disclosure of FRT use in warrant applications, banning arrests based solely on FRT-plus-visual-confirmation, and restricting reliance on searches conducted by unregulated third parties.
Unknown
Developer: Unknown
Country: USA
Sector: Govt - police
Purpose: Identify criminal suspect
Technology: Facial recognition
Issue: Accountability; Accuracy/reliability; Automation bias; Human/civil rights; Transparency
Dec 2019-Jan 2020. An unknown individual commits fraudulent over-the-counter cash withdrawals at SunTrust/Truist bank branches across three Maryland counties.
March 2020. A SunTrust investigator posts suspect images to CrimeDex; a listserv respondent flags Williams via facial recognition.
June 23, 2021. Williams is detained at a Fort Sill military base checkpoint in Lawton, Oklahoma, and arrested on Maryland warrants. Williams spends 23 days in an Oklahoma jail before transport to Maryland.
July-October 2021. Williams is jailed in Montgomery County, Maryland; interrogated by detectives who insist the match is correct.
October 2021. Montgomery County prosecutors drop their charges.
Oct-Dec 2021. Williams is transferred to Prince George's County jail, fighting remaining charges there and in Anne Arundel County.
December 2021. Charges across Montgomery, Prince George's, and Anne Arundel counties are formally dropped, and Williams is released after six months in custody.
April 14, 2026. The ACLU and ACLU of Maryland publicly disclose the incident and file administrative complaints against the three police departments involved.
April 2026. The Washington Post and other media outlets publish investigative reports on the case.
AIAAIC Repository ID: AIAAIC2265