Several AI Chatbots designed for fantasy and sexual role -playing conversations are leaking users’ directions on the network in real time, new research that are seen by cable programs. Some of the filtered data show that people create conversations that detail children’s sexual abuse, according to research.
Conversations with the Chatbots and generative are almost instantaneous: write a message and AI responds. However, if the systems are inadequately configured, this can cause the chats to be exposed. In March, researchers at the UPGuard security firm discovered around 400 AI systems exposed as they scanned the web, looking for erroneous definitions. Of these, 117 IP addresses are directions. Greg Pollock, Director of Research and Information at the UPGUARD, said that the vast majority of these looked like tests, while others contained generic directions related to educational tests or non -sensitive information. “There was a good handful who stood out very different from the others,” says Pollock.
Three of these were role -playing scenarios where people can speak with a wide variety of predefined “characters”, for example, a personality called Neva is described as a 21 -year -old woman living in a university bedroom with three other women and is “timid and often seems sad.” Two of the role configurations were excessively sexual. “Basically used for some kind of sexually explicit role,” says Pollock about the indications on display. “Some of the scenarios involve sex with children.”
For a 24 -hour period, the collected magazine collected by AI systems to analyze the data and try to fix the source of the escape. Pollock says the company collected new data every minute, accumulating about 1,000 filtered indications, including English, Russia, French, German and Spanish.
It was not possible to identify what websites or services are filtering the data, according to Pollock, adding that it is likely to be small instances of AI models that are used, possibly by people more than companies. Pollock says there are no username or personal information of people sending directions to send directions.
Through 952 messages collected by UPGUARD, probably only a vision of how models are used, there are 108 narratives or role -back stages, according to UPGuard’s research. Five of these stages involved children, adds Pollock, including those of up to 7 years.
“LLMs are used to produce mass and then lower the barrier to enter the interaction with fantasies of child sexual abuse,” says Pollock. “There is absolutely no regulation for this, and there seems to be a huge mismatch between the realities of how this technology is used very actively and in which the regulations would be addressed.”
Wired last week reported that a South Korea -based image generator was used to create a childhood abuse generated by AI and exposed thousands of images in an open database. The company behind the website closed the generator after being addressed by Wired. Child protection groups around the world say that the child sexual abuse material generated by AI, which is illegal in many countries, grows rapidly and hinders their work. Charity against the abuse of the United Kingdom has also requested new laws against the Chatbots generative of the AI that they “simulate the offense of sexual communication with a child.”
