Governments are moving fast to wall children off from social media. Australia led in late 2025, and a growing list of nations are following. The legislative logic is protective: reduce harm, limit exposure, give kids their childhoods back. But a new arXiv paper from the same week complicates the premise in ways policymakers seem entirely unaware of.
The Reasoning Gap No Ban Can Fix
A 2026 preprint in arXiv by Rosenbacke, McKee and colleagues, "From Consumption to Reflection: Designing Human-AI Relations for Stable Reasoning," argues that large language models have transformed how humans access information without changing how we reason. The consumption mode that LLMs encourage, fast retrieval, outsourced synthesis, is structurally the same as the social media scroll. Banning TikTok for a 14-year-old while handing them a ChatGPT subscription is rearranging deck chairs. The underlying cognitive infrastructure being eroded is not platform-specific. It is a relationship to friction, uncertainty, and effortful thinking that every frictionless digital system degrades.
Protection That Protects the Wrong Thing
The social media ban conversation is overwhelmingly framed around emotional safety: bullying, body image, predatory content. That is real. But the emerging risk, epistemic dependency on AI-mediated information, is orders of magnitude more structurally significant and almost entirely absent from legislative debate. Marcus Bösch's work on epistemic exhaustion and vibocracy maps exactly onto what the Rosenbacke paper identifies: when affect outpaces argument as the dominant mode of engagement, the capacity for stable reasoning atrophies across a whole population. The countries banning kids from Instagram are treating a symptom. The disease is a media environment, digital and AI-powered alike, that has been optimized for engagement over cognition for thirty years. Children are the canary. The mine is all of ours.