Anthropic: AI May Soon Self-Replicate on Smart Devices
The Facts
- Anthropic said Monday that experiments by its frontier red team suggest small AI systems may be months away from being capable enough to self-replicate on commercial internet-of-things devices, such as smart ovens and off-brand Android TV boxes.
- The company found that systems under 30 billion parameters could complete many of the steps needed for replication but faltered end to end. In simplified tests with some security steps removed, GLM-5.2 and stronger systems achieved full replication.
- Replicating systems would be hard to shut down because there is no central switch-off point, Anthropic said, adding that unlike non-AI malware, they could figure out unfamiliar devices on their own and could target data center processors.
Sources Split
LEFT
CENTER
RIGHT
+1
+1
+1
+4
ESTAB
CRIT
CENTER
PRO
ESTAB
The Spin
Narrative A
Once malware can reason about each machine it lands on, the old patch-and-signature playbook stops working. Free, downloadable systems with the safety rails torn off hand attackers a weapon that costs nearly nothing to scale, since every hijacked device pays for the next attack. Waiting for proof in the wild would be a reckless bet on getting lucky.
Narrative B
Rigging a test network with no passwords, then calling the result a self-replicating AI is pure marketing. A swarm chugging along at a fraction of a token per second and needing days per copy is no match for defenses that have handled worms for decades. Hype like this mostly builds a regulatory moat around closed systems while open weights, the biggest threat to Anthropic and OpenAI's dominance, takes the blame.
Metaculus Prediction
© 2026 Improve the News Foundation.
All rights reserved.
Version 7.18.2