Anthropic says small AI models may soon copy themselves onto TV boxes.
Anthropic's red team published tests on 12 October of open AI models spreading without a data centre. A 9-billion-parameter model set up a working AI model on eight of nine hardware types. Only one open model, Zhipu's GLM-5.2, completed the full copy chain, and only in simplified tests.
Why it mattersSuch a copy would have no central switch to turn off, though it would spread slowly. The team asks device makers for safer defaults and internet providers to watch for unusual traffic.
Read in the app — free, in 9 languages
Related stories
China's top mathematicians formed an alliance to build maths AI.2026-10-13Meta's first frontier model ranks fourth on an independent index.
2026-10-13DeepMind's maths AI passed peer review with nine Erdős proofs.
2026-10-12A model got better by retraining on its own agent teams' best work.
2026-10-12Teams of AI coding agents cost up to five times more for little gain.
2026-10-12