Anthropic published a blog post on October 13 warning that AI models can self-replicate without a data center and hijack everyday connected devices. Reporting since has established that the red team demonstration ran on machines with intentionally exposed Android debug bridge interfaces, and that no small self-replicating model was shown. The discussion now splits between the severity of the warning and the timing, given Anthropic's IPO preparations, with questions about whether it concerns attention, regulatory capture and liability rather than engineering failure.