20% of YouTube's content is now AI-generated. Some of that content is marketed as educational material for children, and it's teaching them to play in traffic and eat toxic food.
Carla Engelbrecht, a former Sesame Street and PBS Kids employee, has been documenting a wave of AI-generated children's videos that present dangerous scenarios as normal. The examples are specific and alarming: videos showing children walking into roads with approaching cars, teaching that "green means right" instead of "go," and depicting babies eating whole grapes (a well-known choking hazard), honey, and raw elderberries - all toxic or dangerous for young children.
"The more content I find, the more horrified I get," Engelbrecht said, calling the videos "downright dangerous."
The Quality Problem Goes Beyond Safety
The danger isn't limited to safety scenarios. AI-generated educational content is also teaching children wrong facts. One state education video featured completely fabricated state names like "Ribio Island," "Conmecticut," "Oklolodia," and "Louggisslia" - the kind of confident nonsense that large language models produce when generating text without grounding in real data.
Dana Suskind, a professor of surgery and pediatrics at the University of Chicago, has joined Engelbrecht in raising the alarm. The concern from children's media specialists is straightforward: young children can't distinguish between accurate and AI-hallucinated content, and the recommendation algorithms treat engagement metrics the same regardless of whether the content is factually correct or physically dangerous.
YouTube's Response So Far
YouTube has deleted some channels with billions of views and is reportedly surveying users on whether content "feels like AI slop." The platform is also testing preview features to counteract clickbait thumbnails and removing flagged videos from reports.
But the scale problem is real. When a fifth of all content on the platform is AI-generated and creators can earn millions from these videos, manual review can't keep up. The incentive structure rewards volume over accuracy, and AI tools make volume nearly free to produce.
This is a different kind of AI safety problem than the ones that dominate the policy conversation. It's not about superintelligence or existential risk. It's about cheap content generation flooding a platform that children use daily, with no reliable system to verify that "educational" actually means educational. For parents relying on YouTube as a learning tool, the practical advice right now is grim: you can't trust the label.