In March 2024, North Dakota police arrested Angela Lipps, a Tennessee woman, for crimes she did not commit. The evidence that put her in handcuffs: an AI facial recognition match. That case—now central to congressional debates over police AI—reveals exactly why the technology remains deeply contested on Capitol Hill.
The encounter began when law enforcement investigating a string of crimes in North Dakota ran suspect images through a facial recognition system. The algorithm returned Lipps as a probable match. Officers obtained an arrest warrant based partly on that output. When she was taken into custody, Lipps had to prove she had never set foot in North Dakota. The charges were eventually dropped, but the damage—legal fees, emotional trauma, a mugshot on file—was already done.
The case exposes a fundamental tension in the push to regulate police use of AI. Law enforcement agencies argue these tools are essential for solving crimes quickly. Federal investigators have used facial recognition in over 1,000 cases since 2015, according to testimony before Congress. The technology can cross-reference millions of images in seconds, something human analysts cannot do. Police unions and prosecutors contend that restricting AI tools would hand criminals an advantage they cannot afford to surrender.
Civil liberties groups counter with a simpler argument: the technology is not accurate enough to base arrests on. Studies consistently show facial recognition performs worse on women and people with darker skin tones. The NIST found error rates can be 100 times higher for certain demographic groups in some algorithms. When a system that is demonstrably biased is used to identify suspects, the result is not efficient justice—it is efficient discrimination, critics argue. They point to Lipps as proof: a woman who became a data point in a broken system.
Congress has struggled to thread this needle. Multiple bills proposing guardrails on police AI have stalled in committee over the past two years. Industry lobbyists argue that regulation would stifle innovation. Civil rights organizations counter that the status quo is already costing innocent people their freedom. Neither side has enough votes to pass comprehensive legislation, leaving departments to set their own rules—or ignore them.
What makes Lipps's case particularly explosive is its timing. Just weeks before her arrest became public, a bipartisan group of senators reintroduced legislation requiring warrants before facial recognition searches and mandating accuracy testing for any AI used in criminal investigations. Supporters hope the concrete evidence of a real wrongful arrest—complete with court records and documented harm—will shift the debate. Opponents warn that overreaction to a single case could cripple tools that have helped solve murders and child exploitation rings.
The North Dakota case now sits in federal court as Lipps pursues a civil rights lawsuit against the department. Her attorney argues the department failed to verify the AI's output before seeking a warrant—basic detective work that would have immediately cleared Lipps. The department disputes this characterization, saying officers followed standard procedures. The outcome of that lawsuit could determine whether police departments face liability for blind faith in algorithmic outputs. For now, Lipps is still waiting for someone to admit that the algorithm was wrong—and to explain how that outcome could have been prevented.