Anthropic Red Team: Small AI Models May Self-Replicate on IoT Devices Within Months
Hidenori8Tanaka · x · 2026-10-13
Anthropic's Frontier Red Team reports that small AI models may be months from self-replicating on commercial IoT devices like smart ovens and off-brand Android TV boxes. In evals, a 9B model ran inference on 8 of 9 hardware types; GLM-5.2 and stronger models completed end-to-end replication in simplified tests. These replicators lack a centralized kill switch and can figure out unfamiliar devices on their own, though doubling takes days. At historical improvement rates, small models may match this capability by year's end.
More from Safety
- Australian PM's Senior Adviser Allegedly Vetted Bipartisan AI Policy Letter, Pocock Concerned — nordicinst · 2026-10-13
- Polymarket Opens Betting Market on Next OpenAI-Reported AI Sandbox Escape — Polymarket · 2026-10-13
- Anthropic says new defenses cut distillation attack data harvesting by 90-99%, cites Zhipu's GLM-5.3 — AnthropicAI · 2026-10-13
- Shared verifiable fix database lets AI agents read and write fixes to each other — Meat_ho · 2026-10-13
- Hacked Danfoss fridges used 'admin' with password set to today's date — J_MaestreVidal · 2026-10-13
- NYT, The Economist, AFP join 80+ publisher coalition against AI content scraping — Polymarket · 2026-10-13