A researcher at Gunma University in Japan has outlined a governance framework for AI systems that analyze street camera footage. The proposal centers on a Fully Monitored Public Space, where dense camera networks track people and vehicles continuously. To keep operators honest, the system introduces an independent recorder that checks each AI output request against a predefined set of legal rules and logs the decision to a tamper-resistant ledger.
The recorder, however, only validates labels. It cannot tell whether an operator's stated purpose, such as finding a missing child, matches what the footage is actually used for. That gap is closed by unannounced audits, which compare recorded declarations with real releases. Penalties, including administrative sanctions and criminal liability, are meant to make misuse more costly than any benefit. The researcher acknowledges there is no single right audit frequency or sanction level, and suggests adjusting them based on experience.
On the public side, the proposal defines acceptance narrowly. People must say they accept the surveillance, give a reason based on no harm or a benefit received, and show no changes in behavior such as route choices or participation in assemblies. An earlier experiment found positive opinions from residents, but it did not measure whether their actions changed. The researcher argues for anonymous surveys and attention to both avoidance and increased feelings of safety, especially among vulnerable groups.
The plan proceeds in three stages, from locating missing children to flagging crimes and eventually using a central AI to identify pedestrians resembling past offenders. The final stage is treated as a stress test, with stopping criteria such as rising chilling effects or false detections. The researcher has not yet discussed the framework with regulators or municipalities, and suggests a small-scale demonstration project before wider deployment.