show index hide index
Harmful Impacts on AI: Thousands of Illicit Files Detected
A shocking investigation by Stanford Internet Observatory experts has uncovered the disturbing presence of at least 1,679 child pornography files in a widespread open data collection called LAION-5B.
This discovery comes at a time when the creation of child pornography by AIs is becoming a growing problem online. The analysis carried out recently by the institution detected that large data archives, intended for the training of visual synthesis systems, include files representing sexual violence against minors. This substantial directory, called LAION-5B, was used to educate the Stable Diffusion AI from the firm Stability AI.
The origin of this set encompasses an overwhelming amount of more than 5 billion visuals and explanatory texts from community platforms and adult content sites. Faced with the performance of AI models in creating convincing visuals from few examples, concern is growing around the potential impact of the presence of such images within LAION-5B. The authors of the study reacted by transmitting the prohibited files to the competent authorities, notably the National Center for Missing & Exploited Children in the USA and the Canadian Center for Child Protection, and they assure that the compromising images are in progress deletion.
Containment and Prevention Actions
As a direct consequence of these alarming discoveries, a LAION correspondent promised an inflexible policy against the presence of illicit content. As a precaution, the non-profit organization that oversees the project has temporarily disabled access to the LAION data collection online for further verification and cleaning.
Meanwhile, a representative from Stability AI claimed that their AI has been packaged with a stripped-down version of LAION-5B. “Filtration systems have been integrated to block problematic commands or results during user interactions with our models,” he specified. These devices are supplemented by identification tools to mark the designed visuals, measures aimed at preventing malicious exploitation of their technology.
