Flock Bolts an Audit Trail Onto 120,000 Cameras, Then Declines to Show the Logic

Flock announced a package of retention, access and auditing changes on Thursday that the company says will curb abuse of its automated license plate reader systems by police customers ^[1]. The changes land on a network of more than 120,000 cameras already installed across the US ^[2], which makes every one of them a retrofit onto hardware, contracts and search habits that predate the policy. The concrete parts are the checkable parts. Flock is now recommending departments keep data for seven days instead of the previous 30 ^[3], a 77 percent cut to the default window ^[4]. Longer retention moves into a new "Evidence Mode" that customers turn on for "exceptional cases" and that requires a specific case number ^[5]. Customers can cap what they share with other agencies by offense type ^[6]. All users will be required to enter case codes to run searches, with administrator override for emergencies such as a missing child, and those overrides get flagged automatically ^[7]. CEO Garrett Langley told The Verge he has changed his mind about what responsibility Flock bears for how law enforcement uses the product ^[8]. Note the verb on retention: recommending. A recommendation configures nothing. The load-bearing control is Audit Assistance, which Flock launched in April, says more than a third of customers have enabled, and will require of everyone by the end of the year ^[9]. It detects abnormal activity, flags it for administrator review, and can lock the flagged user out until an administrator intervenes ^[10]. Flock has not explained in detail how it does that ^[11], and the descriptions on record do not converge. Head of trust and compliance Ashley Haber said it can flag "what may look like an odd search history from a specific user" ^[12]. Co-founder Paige Todd described it flagging a user searching the same plate repeatedly for more than 30 days ^[13]. A customer, Major Patrick Krieg of the Dunwoody, Georgia police department, called it "an algorithm that has notified us of any type of bias" ^[14]. Flock told TechCrunch the feature is not based on machine learning or AI but is a data tool that flags atypical search patterns, offering the example of one plate searched under multiple case codes ^[15]. A heuristic that counts repeated lookups and a system that detects bias are not the same product. What is missing is the arithmetic. Flock has not answered whether it has effectiveness data, what the false positive and false negative rates are, or whether any third party has audited the tool ^[16]. The ACLU's Chad Marlow made the denominator point: without knowing how many officers are misusing the system, there is no way to tell whether the auditing catches 95 percent of violators or 5 percent, and until an independent evaluator looks at it, the tool cannot be distinguished from window dressing ^[17]. The EFF's Cooper Quintin told TechCrunch it may not matter whether it works, because law enforcement will find ways around it ^[18]. Flock has already claimed credit on thin ground once: it acknowledged 15 stalking incidents and said each surfaced because of accountability features deliberately built into the platform ^[19], though some victims learned they had been tracked through HaveIBeenFlocked.com, a site Flock has tried to have taken down ^[20]. The pressure is commercial, not rhetorical. Dozens of communities have cancelled contracts or switched the cameras off, including Los Angeles ^[21]; the LAPD let a three-year deal for 138 pole-mounted cameras lapse while its CIO works on a contract with clear terms on who owns the data ^[22], which Flock attributed to unspecified "misconceptions" ^[23]. Ring ended its partnership after online backlash ^[24]. Watch three things. Whether the year-end mandate actually reaches the up-to-two-thirds of customers who have not enabled Audit Assistance, in the roughly four and a half months remaining ^[25]. Whether the seven-day default becomes a contract term rather than advice.