In this discussion paper, I present an ongoing research-action initiative looking at "Safe City" programs fast spreading across France's urban areas: the Technopolice project. The programs we are documenting consist in research projects as well as on-the-ground experiments where companies from the defence sector or specialised in the operation of network infrastructures, but also global competitors like Huawei and Cisco, can crash-test their "Smart City" security solutions with the support of public authorities, often at the local level and usually without proper legal framework. Applications consist in Big-Data-powered statistics on crime and predictive policing, automatic analysis of video-streams from CCTV cameras (including facial recognition and detection of suspicious behaviours) or of online social networks-technologies for which there is usually a strong IA component. To shed light on these deployments and the public-private assemblages developing them, we have been using FOIA requests and new French laws on "algorithmic transparency". We have engaged in litigation and won a few cases, which in turn is accelerating a process of legalisation. This discussion paper covers the lessons learned as well as the hurdles that our initiative faces, the difficulty in enacting meaningful democratic oversight over these programs as well as on the law enforcement tactics they entrench.