On 25 March 2026, a California jury found Meta and YouTube liable for negligently designing social media platforms that addicted and harmed a young woman. The verdict, which applied product liability doctrine to platform architecture rather than content, was not an isolated event. It was one of three developments that this article examines – the Trojan horse problem, where AI-powered technologies present a benign exterior while concealing architectures designed to exploit users, the rapid shift in the Overton window of regulatory acceptability, as actions once considered unthinkable, such outright bans on social media for minors, multi-billion Euro fines for addictive design, jury verdicts treating algorithms as defective products have materialised in quick succession across multiple jurisdictions, and the implications of these developments for how regulators should approach AI. Drawing on Viktor Shklovsky’s literary concept of ostranenie, or defamiliarisation, the article argues that the central failure in AI regulation is not a failure of law but a failure of vision. Legal instruments are available, but what is missing is the willingness to see AIenabled systems for what they functionally are, rather than what they formally resemble. Regulators, courts, and legislators see familiar objects and apply familiar categories. The harms accumulate in the gap between appearance and reality. The article proposes that regulators must institutionalise a standing disposition to look again at what appears familiar – to ask whether the category that once applied still holds. Three specific demands follow from this: categorical scepticism, functional classification, and pre-emptive scrutiny. The article concludes by examining how this framework applies to India’s developing AI regulatory landscape.