OTTAWA — A recent analysis by a Canadian organization has uncovered child sex abuse images in a dataset used to train artificial intelligence models. The Canadian Centre for Child Protection reported that the dataset contains images of over 120 victims from Canada and the United States, raising significant ethical concerns regarding AI technology development.

Lloyd Richardson, the director of technology at the Winnipeg-based center, stated, "Many of the AI models used to support features in applications and research initiatives have been trained on data that has been collected indiscriminately or in ethically questionable ways." He emphasized that this lack of diligence has allowed known child sexual abuse and exploitation material to appear in these datasets, which he described as largely preventable.

The center's analysis focused on a collection known as Nudenet, which includes tens of thousands of images utilized by researchers to develop AI tools for detecting nudity. These images are sourced from social media and adult websites. The analysis identified approximately 680 images suspected or verified as child sex abuse and exploitation material. Among these, more than 120 images depict victims located in Canada and the U.S., with additional images showing minors engaged in sexually explicit acts.

In response to these findings, the center issued a removal notification to Academic Torrents, a platform used by researchers and universities to access datasets, indicating that the flagged images have been taken down. The center advocates for stricter measures to ensure that datasets distributed to researchers and academics do not contain child sex abuse images and calls for regulatory oversight in AI technology.

This analysis follows a 2023 investigation by Stanford University’s Cyber Policy Centre, which also found child sex abuse images in a dataset used for developing text-to-image AI models. The investigation warned that models trained on this dataset could generate realistic-looking nude images, including those of minors.

Prime Minister Mark Carney has prioritized the development of Canada’s AI capabilities as part of his government’s digital policy. Artificial Intelligence Minister Evan Solomon, the first federal minister to hold this title, is overseeing these efforts. However, he has indicated that the government is not inclined to adopt a regulation-focused approach. Solomon mentioned that the government’s upcoming bill will concentrate on privacy and data issues.

Additionally, Carney’s government has pledged to criminalize the creation of non-consensual sexualized images, commonly referred to as "deepfakes," which are produced using AI tools.