Leaked procurement documents show Berlin’s “behaviour scanners” are designed for cameras, mobile systems and potentially drones

A procurement document exposed in the August cyberattack on Berlin’s administration gives a much clearer picture of the police’s AI-video plans. Separate from the Kottbusser Tor camera project, this specification concerns protection of buildings including the Rotes Rathaus, Abgeordnetenhaus and Interior Ministry. It includes fixed, mobile and temporary cameras, with drone video an optional extension. All could use the same behavioural-alert rules.

Berlin is deliberately avoiding facial recognition and fully autonomous learning. The specification instead requires a human-in-the-loop system: software flags predefined behaviour patterns and a police officer decides what happens next. That design appears partly motivated by Constitutional Court restrictions on more intrusive self-learning surveillance.

The technical test is significant too: detecting a fight or unusual movement from a fixed street camera is one task; applying the same detector to a moving drone looking down from changing angles is considerably harder.

Berlin’s experiment is being designed as a reusable surveillance platform, not a one-off Kotti camera project.