Anthropic conjures another threat that doesn't exist yet to stoke fears of AI apocalypse

Anthropic published experiments showing small AI models can be distributed across low-end devices and, in narrowly controlled tests, propagate copies to vulnerable nodes; this article critiques that work and argues the risk is speculative because the tests used deliberately insecure/custom testbeds, relied on specific models/tools, exhibited slow inference/propagation, and show no evidence of real-world exploitation.