The Netherlands Forensic Institute (NFI) has recently developed a deep learning software that aims to help Dutch police find life-threatening messages sent by suspected criminals.
Indeed, by using a software implant in the encrypted EncroChat phone network, French investigators are able to share live phone data with the Dutch police through a secure computer link. Infiltrating the network has allowed the police to filter around 25 million messages, and predict which could contain a threat to life.
In order to do so, the NFI has used a computer model from 2019 that helped intercept drug-related messages between suspected criminals in large volumes of communications data. Their new deep learning model was then developed similarly to the ‘drug-talk’ software but modified for ‘threat-to-life’ detection and passed on to the police.
The NFI then trained the model’s neural network in generic language comprehension by making it read webpages and newspaper articles, before introducing to messages of suspected criminals, so it could recognize life-threatening messages with key-words. Even though the model is not 100% accurate, its use will be significant to find threats and help the police take action.
The NFI stated that the software was not used in criminal investigations during its development period. The model was ready by April 1st.
It was also stated that there will always be a human making the final decision before intervening.