Legal violations are the order of the day in the virtual world. Children and young people in particular often come into unintentional contact with questionable or even criminal content while surfing. Condat AG has developed an AI system for the Media Authority of North Rhine-Westphalia that is intended to simplify, accelerate and improve media supervision on the Internet.

Depictions of violence, incitement of the people, the use of anti-constitutional symbols or freely accessible pornography: the list of websites or social media offerings with illegal content is long and constantly growing. This brings employees of the state media authorities to the limit of their workload. Manually combing the virtual space for critical content simply cannot be done.

KIVI is the name of a software program designed to ensure greater security on the network. It was developed by the Berlin-based company Condat AG and was first used in 2020 by the Media Authority of North Rhine-Westphalia - initially as a prototype, and since 2021 in productive operation. Whether World Wide Web, YouTube, Twitter, Telegram or other social media platforms: KIVI can automatically search more than 10,000 pages every day. What would take a human days or weeks, the AI tool can do in a fraction of a second.

Ongoing improvement through daily feedback

KIVI learns from image and text examples that are actively fed in and have been assessed as violations in the past. In addition, the software receives daily feedback from employees of the state media authority as to whether a suspicion it has found has been confirmed or not. In general, the more authorities that work with this software, the more efficiently the AI tool can learn. Currently, KIVI is only trained for German-language text content. In the meantime, it is also being used in some other state media authorities in Germany.