Flock, a maker of AI-powered surveillance cameras used widely by US police departments, has faced growing backlash over the past year as departments cancel contracts and city councils vote to remove its systems. This followed controversies including a Texas deputy searching more than 83,000 cameras for a woman who had had an abortion, and Illinois finding that Flock had let federal immigration agents access state camera data unlawfully. WIRED's examination of Flock's code found that while the company introduced new safeguards in August, these guardrails may discourage misuse and record it, but do not actually prevent it, leaving much of what governs a search hidden on Flock's own servers.

Flock's reforms, due to become mandatory by year's end, include shorter default data retention, mandatory case codes for searches, and automated auditing. However, experts say Flock's AI is tasked with reliably screening search descriptions at a scale no technology can perform accurately, and no outside party can measure its error rate. Notably, of the sensitive categories the system screens against, including race and religion, only "political, social, and cultural expression" fails to actually block a search, despite this being an area courts have historically given the strongest free-speech protection. Flock told WIRED its tools include "clear safeguards" but did not explain how its AI model was built or how moderation decisions are made.