Don't be fooled. the US is regulating AI – just not the way you think (www.theguardian.com)

Don’t be fooled by the “deregulation” rhetoric: the US is actively regulating AI — but at the infrastructure level rather than at visible consumer applications. While politicians and administrations publicly warn against smothering chatbots and image generators with rules, Washington has been imposing export controls on advanced AI chips, restricting access to supercomputer-class hardware and even targeting model weights. Both recent administrations have shaped where compute and models flow (Biden’s chip export limits to China; Trump-era outreach to partners like the UAE), signaling a strategic focus on the AI stack’s building blocks—hardware, datacenter capacity and core model artifacts—rather than surface-level apps. For AI/ML practitioners and policymakers this shift matters: controlling chips and model weights directly constrains who can train state-of-the-art models, where experimentation happens, and how quickly capabilities diffuse globally. It reframes regulation from ethics- and harm-centered rules (the EU’s application-focused approach) to national-security-driven controls, with China pursuing its own model and content restrictions. The authors argue a hybrid, stack-aware governance—combining societal safeguards and strategic export/trade policy—works best and call for greater transparency about these low-level interventions. Recognizing regulation’s true locus is essential for realistic global coordination, research planning, and responsible capability diffusion.