Researchers warn US politics is repeating its ChatGPT mistake with world models

The warning isn't just about world models in the abstract — it's about embodied AI as a distinct strategic category that US policy has yet to treat separately from LLMs, meaning the governance and funding frameworks built around language models may be structurally mismatched for what comes next.

MIT Technology Review's piece on the history of robotics learning (story 6) is the clearest prior context here: it traced how roboticists have repeatedly overpromised on humanoid systems while delivering narrow industrial ones, which is precisely the pattern researchers are now warning policymakers not to repeat at the geopolitical level. The UK's $675 million sovereign AI fund (story 7) shows at least one government is moving toward capability-specific investment, though that announcement was framed around software and startups rather than embodied systems. Anthropic's cybersecurity model release (story 2) signals that some AI labs are actively courting government alignment on national security use cases, but world models sit outside that conversation almost entirely right now.

Watch whether the next US executive AI policy action or Congressional hearing explicitly names world models or embodied AI as a distinct funding category. If it doesn't appear in any formal policy language within the next six months, the researchers' warning will have gone unheeded in exactly the way they predicted.