WASHINGTON – U.S. Senators John Cornyn (R-TX) and Andy Kim (D-NJ) today introduced the Preventing Recurring Online Abuse of Children Through Intentional Vetting of Artificial Intelligence (PROACTIV AI) Data Act, which would encourage artificial intelligence (AI) developers to identify, remove, and report known child sexual abuse material (CSAM) from the datasets they compile or obtain for use in training AI models to help proactively stop AI image generators from creating child pornography:

“Modern predators are exploiting advances in AI to develop new AI-generated child sexual abuse material, and technology companies are often unwittingly giving them the tools to do so,” said Sen. Cornyn. “By encouraging tech companies to proactively screen their datasets to remove and report explicit images of children, this legislation would mitigate the risk of AI platforms unintentionally enabling the creation of new content depicting child sexual abuse and better safeguard children online.”

“As we develop AI models, it is important that we establish critical protections to look out for the most vulnerable in digital spaces,” said Sen. Kim. “This bill is an opportunity for Congress and AI developers to take an important step forward together and implement the necessary safeguards to keep our children safe from future misuse or exploitation.”

Background:

Foundational AI models require large amounts of data for training purposes. Given the size of these datasets and the amount of compute needed to work through them during training, many companies do not screen content for harmful materials. Recognizing the potential for child sexual abuse material (CSAM) to unintentionally work its way into these datasets given the methods of collection, researchers at Stanford University screened multiple data sets, including the LAION-5B data set used to train many of the leading image-generating models. Their work identified more than 3,000 data entries of likely CSAM in the dataset. This is troubling, not just because of the accessibility of known CSAM, but because many products and tools now have content from which to draw if asked to generate new CSAM by users. Left unaddressed, foundation models can be corrupted by CSAM data and make it possible for them to be coopted by predators.

According to the National Center for Missing & Exploited Children, AI-generated material has proliferated at an alarming rate in the past year with nearly half a million incidents of AI-related CSAM reported in the first half of the year, compared to fewer than 70,000 reported for all of 2024.

The PROACTIV AI Data Act would:

• Direct the National Institute of Standards and Technology (NIST) to issue voluntary best practices for AI developers to screen their datasets used for training AI models for known CSAM;
• Direct the National Science Foundation to support research into innovative methods and technologies for identifying, removing, and reporting CSAM from datasets;
• And provide limited liability protection to AI companies that follow the new best practices so those acting in good faith aren’t legally punished for accidentally handling CSAM via automated data crawlers on the internet.

This legislation builds on First Lady Melania Trump’s efforts to combat the rise of AI deepfake pornography and keep children safe online.