Google, Meta, OpenAI, Microsoft and Amazon have committed to review their AI training data and remove any material that may refer to child sexual abuse (CSAM).
CSAM has already been found in educational materials of artificial intelligence models. A wide range of companies finally agreed to fight this phenomenon. Google, Meta, OpenAI, Microsoft, Amazon, Anthropic, Stability AI and others have signed a new set of principles aimed at limiting the spread of CSAM. They promise to ensure that their training datasets are CSAM-free, avoid datasets with a high risk of including CSAM, and remove CSAM images or references to CSAM from data sources. The companies also commit to “stress testing” AI models to ensure they don't generate any images depicting sexual abuse, and only release models if they've been assessed for child safety.
In a blog post, Google said that in addition to following the guidelines, the company has also increased advertising grants to the US National Center for Missing and Exploited Children (NCMEC) to promote its initiatives.
CSAM was detected, in particular, in the LAION-5B dataset on which Stability AI trained its Stable Diffusion AI model.