Flock Safety's latest software suite is empowering police with advanced search tools that go far beyond traditional license plate recognition. These capabilities, which include an AI-powered watchlist, allow officers to define an area and run continuous, automated searches for individuals matching written descriptions. Over the past year, **Flock Safety** has faced increasing backlash, leading to contract cancellations and votes to remove their cameras in various municipalities. Congressional members have also voiced concerns following incidents such as a Texas deputy's search for a woman who had an abortion, and Illinois found the company in violation of state law for allowing federal immigration agents access to state camera data. In response, **Flock Safety** introduced a package of changes in August, including shortened default data retention periods, mandatory case codes for searches, and automated auditing to flag misuse. While these changes are set to be mandatory by year-end, analysis suggests that these guardrails may discourage some abusive uses and record them, but crucially, they may not prevent them entirely. ### The Accuracy Dilemma of AI-Powered Searches Experts studying automated screening and police technology highlight a critical issue: **Flock Safety**'s AI is tasked with a job no technology reliably performs. The system's decisions regarding search parameters are made out of sight of the departments using it, and the accuracy of such large-scale screening is questionable. Furthermore, external parties cannot measure the frequency or direction of its inaccuracies. **Flock Safety** itself warns officers that results may be inaccurate or incomplete and should not be solely relied upon, placing the "risk of any inaccuracies" on the officer conducting the search. One particularly contentious aspect is the system's handling of sensitive content categories. While it scores an officer's written description against categories like race and religion, political, social, and cultural expression is the *only* category that will not stop a search. **Tom Bowman**, policy counsel with the **Center for Democracy and Technology's security surveillance project**, notes that political and cultural expression is one of the most protected areas under the First Amendment, making **Flock Safety**'s approach here particularly concerning. When questioned, **Flock Safety** stated its tools are built with safeguards and that queries are evaluated against content policies. However, the company did not disclose details about its AI model's construction, operating instructions, or the criteria separating one verdict from another. ### How the Tools Operate **Flock Safety**'s interface includes various tools, each with distinct functionalities. For instance, a text-to-image search produces a statistical ranking of footage based on how well it matches the typed description. Every captured image and typed description is converted into numerical data, which the model then compares. A feature called **Smart Sort** allows officers to approve or reject images, refining search results by moving similar approved footage to the top. While vehicle searches offer structured menus for attributes like color and make, person searches, known as **FreeForm**, accept only written descriptions. This **FreeForm** capability, which **Flock Safety** states it has sold for years, allows officers to search for "clothing, colors, objects, or other case-relevant details" on video cameras. Analysis of **Flock Safety**'s user interface code reveals that, for now, departments can configure some searches without requiring a reason or case number, and can even disable the automated system for flagging suspicious use. **Flock Safety** asserts these controls will be mandatory by year-end, and that search activity is recorded in audit logs. However, it remains unclear whether anything on their servers would prevent a search submitted without a reason or case number. ### A History of Misuse and Abuse Concerns about **Flock Safety**'s tools are amplified by a history of law enforcement personnel misusing similar investigative databases. According to The Washington Post, at least 50 officers in the U.S. have recently been charged with or accused of misusing license plate readers, with 46 of those cases involving **Flock Safety**. More than half of these alleged abuses targeted wives, girlfriends, ex-partners, or individuals the officers wanted to meet. Past incidents of database abuse include: * A Milwaukee officer searching for a woman he was dating 124 times and her former partner 55 times, logging each as an "investigation." * A Georgia police chief running his ex-girlfriend and her daughter 600 times, leading to charges and his eventual suicide. * A Kansas chief losing his badge after running his ex 164 times and her boyfriend 64 times. * A North Carolina officer running her boyfriend's ex-wife 31 times, filing 29 as traffic infractions. This pattern of misuse is not new. A 2016 Associated Press investigation found over 325 instances of officers running database searches on exes and attractive individuals between 2013 and 2015, resulting in firings, suspensions, or forced resignations. More recently, investigations by **WIRED** in 2023 and this year uncovered hundreds of allegations against **Immigration and Customs Enforcement (ICE)** and **Customs and Border Protection (CBP)** employees for misusing sensitive government databases to track individuals for personal reasons.  **Flock Safety**'s software includes two notification systems aimed at human users. One is the watchlist based on police search queries, which allows an officer to define an area on a map and have the system alert them to anyone matching a description. The efficacy and ethical implications of such broad, AI-driven surveillance tools continue to be a subject of intense debate and scrutiny.