The American company Flock Safety, known for its surveillance camera networks, is developing a new AI tool called OS Investigate. This system has the potential to radically change how law enforcement approaches investigations: it allows officers to locate drivers based on their routes and behavioral patterns, without having either the suspect's name or their vehicle's license plate number. I managed to analyze the technical details that reveal the scale of this development's capabilities.

How OS Investigate works

During a technical analysis of the code posted on the developer's website, more than 450 files were discovered that reveal the system's interface. This data makes it possible to reconstruct the tool's operational logic without direct access to police accounts. A key feature is the presence of 69 ready-made AI query templates that an officer can adapt to a specific task or create their own from scratch.

The system integrates with a vast array of data: street camera footage, police reports, 911 call logs, arrest records, and commercial databases containing phone numbers, email addresses, dates of birth, and information about relatives. Of particular interest is the fact that 19 of the 69 templates are aimed at identifying behavioral patterns, and 14 do not require a name, license plate number, or vehicle description at all — simply specifying an area, time window, and type of activity is sufficient.

Searching for "connections" between vehicles

Even more powerful is the "connections" analysis feature. The algorithm tracks which vehicles appear at the same cameras as the target car within a two-minute interval. If such a match is recorded at least three times and the confidence level exceeds a set threshold, the vehicle is added to a list of potential associations — the system can return up to 20 such results. This effectively allows building movement graphs and identifying accomplices or family members without any initial data.

Flock representatives claim that the cameras do not identify the driver, and data is stored for only 30 days. However, OS Investigate, according to the developers, is a separate product that works with information to which police already have legal access. Notably, the request justification form has no strict requirements: even if an agency requires a case number, entering at least three characters is sufficient. This raises serious questions about oversight of the system's use.

Context and risks

The launch of the new tool comes amid growing criticism of Flock. The company's network covers more than 6,000 communities in the United States and processes around 20 billion license plate reads monthly. The Justice Institute has documented dozens of cases of abuse, including surveillance of former partners. Although the company insists such incidents are isolated, pointing to logging of all actions, the addition of AI features with vague query criteria only heightens privacy experts' concerns.

My analysis: the introduction of predictive algorithms into law enforcement practice without strict regulatory constraints is a path toward creating a system of total surveillance. The ability to "link" vehicles and search for people by behavior rather than identifiers blurs the line between prevention and unjustified intrusion into private life. The industry and lawmakers urgently need to develop clear standards, otherwise technology will outpace the law.