Skip to main content
Incident intelligence/SS-IR-021CASE FILE OPEN
Symbolic editorial illustration for SS-IR-021SERVANTSTACK // INCIDENT INTELLIGENCEFORENSIC IMAGE // VERIFIED FRAME
SS-IR-021 // INCIDENT REPORTReported

Detroit Police

Face Recognition Misidentified an Innocent Man and Got Him Arrested in His Own Driveway

EXECUTIVE BRIEF

In January 2020, Detroit police arrested Robert Williams outside his home in Farmington Hills, in front of his wife and two young daughters, and held him for roughly 30 hours in a crowded cell.

SURVEILLANCE & POLICING
FAILURE CHAINTRACE COMPLETE
  1. 01TRIGGERIn January 2020, Detroit police arrested Robert Williams outside his home in Farmington Hills, in front of his wife…
  2. 02MACHINE ACTIONDecision system
  3. 03MISSING GATENamed SME review and decision audit trail
  4. 04IMPACTHuman welfare
01 // INCIDENT SUMMARY

The short version

In January 2020, Detroit police arrested Robert Williams outside his home in Farmington Hills, in front of his wife and two young daughters, and held him for roughly 30 hours in a crowded cell.

02 // KEY FACTS

Case telemetry

INCIDENT
SS-IR-021
DATE
January 2020
SYSTEM
Detroit Police
LOCATION / SCOPE
Detroit, Michigan, United States
EVIDENCE
Reported
AI ROLE
Decision system
HARM
Human welfare
SOURCES
3 cited records
03ENTRY POINT // WHAT HAPPENED

The event

In January 2020, Detroit police arrested Robert Williams outside his home in Farmington Hills, in front of his wife and two young daughters, and held him for roughly 30 hours in a crowded cell. The case against him rested entirely on a facial recognition "match." Investigating a 2018 theft of five watches (about $3,800 in merchandise) from a downtown Detroit Shinola store, a detective sent a blurry, low-quality still pulled from the store's surveillance video to Michigan State Police, who ran it through a DataWorks Plus face recognition system. The algorithm returned Williams' expired driver's license photo as a candidate. He was the first person publicly known to be wrongfully arrested in the United States because of a face recognition error. The charges were dismissed; the ACLU sued the City of Detroit in April 2021, and in June 2024 the city settled for $300,000 and agreed to the nation's strongest police limits on the technology.

04CAUSAL TRACE // AI'S ACTUAL ROLE

What the machine did

The face recognition system did exactly one thing: it returned a ranked list of candidate faces from a grainy, partial image and surfaced Williams as a probable match. It was never designed to confirm identity, and its own vendor warns that results are investigative leads, not probable cause. There was no human SME gate to test the lead before it became an arrest. No one corroborated the match against an alibi, a credit card trail, location data, or even a careful side-by-side look (Williams later noted the suspect in the photo looked nothing like him). The investigator treated a statistical similarity score as ground truth, the lead was laundered through a quick photo lineup, and machine output went straight to handcuffs. The model produced a guess at machine confidence; the institution promoted it to a fact with zero validation in between.

Decision systemAutomation was a causal participant—not a decorative label for the system around it.
05BLAST RADIUS // CONSEQUENCES

Where the failure landed

An innocent man was arrested in front of his children, fingerprinted, photographed, DNA-swabbed, and jailed for about 30 hours over a crime he had nothing to do with. The charges were eventually dropped, but the arrest stayed on record and the trauma did not wash off. The case became the canonical example of face recognition's civil rights failure mode, especially its documented higher error rates on Black faces, and was soon joined by similar wrongful arrests (including Nijeer Parks in New Jersey and Porcha Woodruff in Detroit). In June 2024 Detroit paid Williams $300,000 and accepted binding reforms: no arrests based solely on a face recognition result or on a lineup that flows directly from one, mandatory corroborating evidence, officer training, and an audit of every face recognition case since 2017.

06 // EVIDENCE STATUS

Reported

Documented in the cited public record. Follow the sources for the precise evidentiary posture.

SOURCE RECORD UPDATED 2026-07-09

07 // SOURCE LEDGER

3 cited records

  1. 01
  2. 02
  3. 03
08CONTROL FAILURE // MISSING GOVERNANCE

Named SME review and decision audit trail

The failure pattern in this case: Automated judgment without accountable review.

09INTERVENTION POINT // HUMAN IN THE MIDDLE

The moment the path could change

A qualified reviewer tests the basis, context, and disparate impact before the decision reaches a person.

AI PROPOSES→HUMAN OWNS THE DECISION→SYSTEM EXECUTES
10CONTROL DEPLOYMENT // AUTHORITYGATE

SME routing · decision audit trail

The AuthorityGate Operational Resilience framework treats an algorithmic identity "match" as an unverified lead that cannot advance to any consequential action until a qualified human SME signs off on independent corroboration. A face recognition candidate would enter a change-validation gate that blocks the next step (warrant request, lineup, arrest) until a reviewer attests, on the record, that the match is supported by evidence the algorithm did not produce: image-quality sufficiency for the probe photo, an alibi check, location or transaction data, and a documented same-person determination by a trained examiner. Crucially, the gate forbids "circular corroboration" -- a lineup or witness ID seeded by the algorithm's own output does not count as independent. The probe-image quality itself is gated: a blurry, partial still below a defined resolution and pose threshold is rejected before a search is even run, so a low-confidence guess never becomes the spine of a case. No SME attestation, no escalation; the lead simply cannot leave the investigative tier.

RELEVANT KEYSTONE CONTROLHuman-in-the-Loop ValidationHow high-risk actions route to a named subject-matter expert who owns the go or no-go decision.
12 // THE ALTERNATIVE

Autonomy is a design choice.

See the operating model that keeps AI useful while preserving human authority at consequential moments.

Compare AgenticAI and AugmentedAI →