WIRED has reconstructed the interface of Flock Safety’s AI-assisted search tools, which can help US police continuously scan cameras for people matching a written description. The central concern is blunt: some safeguards appear designed to warn and record officers, not necessarily stop a sensitive search.
Watch Desk analysis
What happened
Published on 3 September, WIRED’s investigation draws on code Flock sends to a user’s browser, interviews with specialists and a detailed response from the company. The reconstruction covers the controls officers see and the requests the interface expects, but not Flock’s server-side model, live camera feeds, actual results or individual department settings.
Among the tools described is a watchlist that lets an officer draw an area on a map and receive alerts when cameras detect someone the model considers a match for a written description. A separate alert can notify officers whenever a person enters a selected part of one camera’s view.
Why it matters
A continuous watchlist changes a camera archive from something police search after an event into a system that can actively look for people across an area. When the matching model is opaque and its errors cannot be measured externally, an alert can look far more certain than the machinery beneath it deserves.
Audit logs and case codes can deter misuse and make later scrutiny possible. They are not the same as prevention. That distinction matters when officers have previously been accused of using police databases to track partners, former partners and other people for reasons unrelated to police work.
Our read
Flock’s safeguards sound more substantial when described as blocks, warnings and audits. WIRED’s reconstruction suggests the verbs matter: a warning that still offers a continue button is oversight with an escape hatch.
Police departments considering these tools should demand independent accuracy testing, compulsory case-linked searches and rules that prevent prohibited queries before deployment. Procurement teams should also establish who reviews alerts, how quickly and with what consequences. A log nobody checks is merely a very tidy alibi.
WittyWires has not independently inspected the code or Flock’s server-side systems. Flock told WIRED that prohibited attributes are blocked, searches are evaluated against content policies and its tools include clear safeguards, but it did not explain the model or the thresholds separating verdicts.
What to watch
- Whether Flock makes case codes and automated misuse detection mandatory by the end of 2026.
- Whether independent auditors receive enough access to measure false matches and uneven error rates.
- How departments govern searches involving protests, clothing, symbols and other protected expression.
- Whether Flock clarifies retention and internal access rules for flagged search records.
Discussion spark: Should an AI police-search tool ever let an officer override a warning involving protected political or cultural expression?
Sources and evidence
- Source update (3 September 2026, 10:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.