Document Category: Flock

This motion argues that Flock’s ALPR and Vehicle Fingerprint® systems are secretive, unregulated, and error-prone artificial intelligence tools whose reliability cannot be assessed without robust discovery into how they function, how they were developed and trained, and what their error rates are (pp. 4-7). It explains that machine-learning “hallucinations,” environmental conditions (weather, lighting, camera noise), and reliance on flawed databases and private HotLists undermine the accuracy of Flock outputs (pp. 4–7). Drawing on Maryland Rule 4-262, Brady/Giglio, and Crane v. Kentucky, the motion contends that Flock operates like an uncorroborated confidential informant or other similar technologies like ShotSpotter and Facial Recognition and that the accused is therefore constitutionally entitled to underlying data, logs, policies, audits, and technical documentation so they can confront and impeach Flock-derived evidence and present a complete defense (pp. 8–14).

This motion seeks to invalidate a warrant and suppress all resulting evidence where the warrant rests solely on Flock data. It explains that Flock’s machine-learning model generates conclusions based on a non-public AI model with an unknown error rate, is vulnerable to hallucinations, and is highly sensitive to weather, lighting, noise, and partial plate captures, with studies showing accuracy can fall to “near zero” under certain conditions (pp.1-6). Because Flock functions like an uncorroborated anonymous informant, the motion argues that Flock data cannot establish sufficient veracity for probable cause and lacks particularity (pp. 7-12). It further contends that omissions about Flock’s known flaws defeat the good-faith exception (pp. 12-15).