DarkBert AI is a new kind of artificial intelligence: rather than collecting its data on the classic Web, it has been trained on the Dark Web. His goal ? Better understand the codes and uses of the underground network.
If artificial intelligences usually train by collecting millions of data and images on the Web, there is one that stands out completely from its “competitors”. Because ChatGPT, Google Bard or Bing Chat are missing a bowstring. baptized Dark Bert AIthe artificial intelligence in question conducted its research on the Dark Web.
Coming from a project launched in 2019 and named RoBERTa, artificial intelligence thus harvested information from hackers and cybercriminals of any hair.
DarkBert AI is an artificial intelligence like no other
In order to train DarkBert AI, its designers scoured the Dark Web using the Tor network. They thus collected millions of data, which they then filtered, which enabled them to develop a huge database. As a result, they were able to improve the LLM (Large Language Model) of RoBERTa, so that the AI is able to interpret the different codes of the Dark Web.
A long-term work and carried out in English. The designers of this new kind of AI insist on the fact that 90% of the content exchanged on the Dark Web is in the language of Shakespeare. The team in charge of DarkBert AI has also taken care not to reference any illegal content. She explains her approach as follows: “When exploring the Dark Web, we are careful not to expose ourselves to content that should not be accessible. For example, illicit pornographic content (like child pornography) is easily found on the Dark Web. However, our automated web crawler removes all non-text media and only stores raw text data. In doing so, we ensure that we are not exposing ourselves to sensitive media that is potentially illegal.”
The purpose of such AI is not to help hackers develop new attacks or to allow cybercriminals to better hide themselves from law enforcement. On the contrary: if DarkBert AI was created, it is for provide a way for authorities to better understand the Dark Net and for security experts to better protect ordinary users.
Like any other artificial intelligence, DarkBert AI is still in the learning phase. The project is not available for testing at the moment.
Source : Arxiv
