Matériel d'exploitation sexuelle d'enfants généré par l'IA au Canada

The proliferation of generative AI image models has created an emerging and documented concern for child safety: the ability to generate photorealistic child sexual abuse material (CSAM) using text-to-image AI tools. The Canadian Centre for Child Protection (C3P), which operates Cybertip.ca and Project Arachnid, has identified AI-generated CSAM as an escalating concern, with reports of synthetic abuse imagery increasing from 2023 onward (Canadian Centre for Child Protection, 2024).

Open-source image generation models can be fine-tuned or prompted to produce exploitative imagery of children. Unlike traditional CSAM, which documents actual abuse, AI-generated material can be produced at scale without requiring access to a victim — but child protection organizations warn it normalizes the sexualization of children, can be used to groom real victims, and threatens to overwhelm the detection infrastructure that organizations like C3P have built over decades (Canadian Centre for Child Protection, 2024). Hash-based detection systems like PhotoDNA, designed to match known CSAM images, are not designed to identify novel AI-generated content.

Canadian law addresses CSAM through Criminal Code provisions that cover visual representations depicting minors in sexual activity, which is widely interpreted as covering synthetic material, though definitive appellate-level interpretation remains limited. Prosecution of AI-generated CSAM cases is still in early stages — Steven Larouche of Sherbrooke, Quebec was sentenced in April 2023 to a total of eight years — approximately three and a half years for creating at least seven deepfake child pornography videos, and four and a half years for possessing over 545,000 files of child sexual abuse material (Canadian Centre for Child Protection, 2024). The presiding judge described it as the first case in Canada involving deepfakes of child sexual exploitation. The volume of synthetic material risks straining law enforcement resources, and distinguishing AI-generated from real imagery is increasingly difficult.

Canadian law enforcement, including the RCMP's National Child Exploitation Coordination Centre (NCECC), and child protection organizations are calling for coordinated action: stronger model-level safeguards from AI developers, updated legal frameworks, new detection technologies, and international cooperation to address a transnational problem that accessible generative AI tools make worse (CBC News, 2024; Public Safety Canada, 2004).