Anthropic red team: AI models may soon self-replicate on $55 TV boxes, no datacenter needed
alex_verem · x · 2026-10-16
Anthropic's Frontier Red Team reports small AI models may be months from self-replicating on commercial IoT devices like smart ovens and off-brand Android TV boxes.
- The team networked 15 $55 TV boxes and ran a 26B open model across them at 0.39 tokens/s—slow, but proof capable models can run outside datacenters
- A 9B model got inference running on 8 of 9 hardware types tested (Jetson Nano, Macs, RTX 4090s); 30B-and-under models can already complete most individual replication steps but fail composing them end to end
- In simplified evaluations, GLM-5.2 and stronger models achieved end-to-end replication; at historical improvement rates (13.7 points/year on Epoch AI's index), sub-30B models should match this by end of 2026
- Unlike classic malware, such a worm has no central kill switch and can handle unfamiliar devices, though it's slow (days-level doubling time), computationally modest, and detectable by ISPs
- Mitigations include banning guessable default passwords, as the UK has done since 2024
More from Safety
- Gig workers face scans alongside 16 trackers: how real is their consent? — AntonioCasilli · 2026-10-16
- Dev trains his own PII detector with decoy-entity training samples to beat existing tools — HankYeomans · 2026-10-16
- Anthropic grants METR inside access to audit its AI models and safety practices — The Decoder · 2026-10-16
- Agent actions need checks outside the model: NVIDIA demos lean on guardrails amid wave of real agent incidents — Creamy-And-Crowded · 2026-10-16
- Vitalik builds "paranoid mode" agent wrapper with Tor, zkAPI and local prompt rewriting — DavideCrapis · 2026-10-16
- Security team demos Signal 1-click exploit technique at Hexacon Paris — moyix · 2026-10-16