Technology companies including Google, Meta, OpenAI, Microsoft, and Amazon today committed to reviewing their AI training data for child sexual abuse material (CSAM) and removing its use in future models.
The companies have signed a new set of principles aimed at limiting the spread of CSAM. They promise to ensure that their training datasets do not contain CSAM, avoid high-risk datasets containing CSAM, and remove CSAM images or links to CSAM from their data sources. The companies also promise to “stress test” their AI models to ensure they do not produce CSAM images, and only release models once they have been assessed for child safety.
Other signatories include Anthropic, Civitai, Metaphysic, Mistral AI, and Stability AI.
Generative AI is contributing to growing concerns about deepfake images, including the proliferation of fake CSAM photos online. Researchers at Stanford University published a report in December that found that common datasets used to train some AI models contain links to his CSAM images. Did. Researchers also found that the tip line run by the National Center for Missing and Exploited Children (NCMEC), which is already struggling to handle the amount of CSAM content reported, is rapidly being overwhelmed by AI-generated CSAM images. I also discovered that something is wrong.
Thorn, a child abuse prevention nonprofit that helped develop the “All Tech Is Human” principles, says AI image generation will hinder efforts to identify victims, create further demand for CSAM, and reduce children to victimization and re-victimization. It enables new ways to victimize and makes it easier to find information on how to share problematic content.
Google said in a blog post that in addition to adhering to the principles, it has also increased advertising subsidies to NCMEC to further its efforts. Susan Jasper, Google's vice president of trust and safety solutions, said in a post that supporting campaigns like this increases public awareness and provides tools to identify and report fraud. Ta.
