The power AI is providing for the IoT deployment has become a game changer, being one of the drivers of IoT. Offering a multitude of advantages based on the capacity to automate services, management, and maintenance of IoT devices, but current centralized solutions present certain vulnerabilities such as security, data processing, and real-time data to information. Resulting in inaccurate data feeding to AI, as well as overloading the bandwidth with unnecessary data, consequently latency, cyberattacks, and Big Data arises as constant problems. An example can be seen within smart city networks like a traffic intersection where cameras are generating data on a 24/7 basis that is moved constantly to server centers. This situation presents issues because the majority of this Big Data generation isn’t relevant for traffic flow, in general, doesn’t present incidents. Therefore, the AI processing of data costs and the bandwidth capacity used becomes, instead of a solution, a problem. To solve it we will present a decentralized AI data to information refinement, a process that works on premises, filtering irrelevant IoT data to only process incident data. In this way, bandwidth is freed and first responders receive real-time IoT computation to actuate much quicker than centralized solutions. In addition, being decentralized makes cyberattacks practically implausible.