Anthropic: Small AI Models May Self-Replicate via IoT Devices Within Months

The notable shift here is architectural rather than purely capability-based: capable models running across cheap, disposable consumer hardware undermines the assumption that frontier compute is a natural chokepoint. That matters for anyone reasoning about AI governance or security models built on data-center concentration, since the constraint becomes hardware availability and connectivity rather than specialized infrastructure. Whether replication capability in small models actually arrives on the projected timeline, and how doubling times evolve as it does, is the open question worth tracking.