SignalCapabilitySG-0032
Anthropic's red team says small AI models may be months from self-replicating on cheap smart devices like TVs
Small models already do many replication steps but fail to chain them; GLM-5.2 and stronger models did the whole cycle in simplified tests. At their historical rate of improvement, small models would reach that level by the end of 2026. The team also ran a 26-billion-parameter model split across 15 off-brand Android TV boxes costing about $55 each. This is a projection from tests that made the hacking step trivial, and the team says no full rogue replicator exists yet.
Why it matters
A replicator spread across many cheap devices would have no central point to switch off, and unlike scripted malware it could work out unfamiliar hardware on its own. Small open models running on hijacked home devices sit outside any lab's controls, so stopping one would fall to network operators and device makers.
Sources
Read the reporting
2 sources. Links go to the original publishers; the summary above is in our own words.
- primaryAI models may be capable of self-replicating without access to a datacenterAnthropic · Oct. 12, 2026anthropic.com/research/self-replicating-ai
- newsAnthropic Red Team Says Small AI Models May Soon Replicate on IoT DevicesUnite.AI · Oct. 12, 2026unite.ai/anthropic-red-team-says-small-ai-models-may-soon-replicate-on-iot-devi…
Signals
More from the news desk
Cite and share
Use this signal
Citation
Paperclip Index. “Anthropic's red team says small AI models may be months from self-replicating on cheap smart devices like TVs.” Signal SG-0032. Reported Oct. 12, 2026; updated Oct. 12, 2026. https://paperclipindex.com/signal/SG-0032