2026-10-12
Anthropic tests show small AI models nearing self-replication on IoT devices
Anthropic's Frontier Red Team says a 9B-parameter model ran inference on 8 of 9 tested IoT hardware types, and reports that GLM-5.2 and larger models completed end-to-end self-replication in simplified evals, per anthropic.com (Oct 12, 2026) — its own, unverified figures.
What else the source says
- 9B-parameter model achieved inference on 8 of 9 tested IoT hardware types
- Models of 30B parameters or fewer can already do individual replication steps but fail end-to-end
- GLM-5.2 and stronger models completed full end-to-end replication in simplified evals
- Replicator's doubling time estimated at days, per Anthropic
- Small models projected to match end-to-end capability by end of 2026 if trend continues
Why it matters
This suggests malware could soon run without centralized datacenters.
Source. anthropic.com
TRACE
- First seen Oct 12, 2026, 12:16 PM PT
- Published Oct 12, 2026, 4:17 PM PT, 4 h after first seen
- Source anthropic.com (Company source)