Anthropic: Small AI models may gain the ability to self-replicate via smart devices within months.

PANews October 13 news: According to research published on Anthropic's official website, its Frontier Red Team believes that small AI models may be able to achieve self-replication within just a few months without relying on data centers, using cheap IoT devices. In testing, a 9B-parameter model successfully set up inference services on 8 out of 9 types of hardware; Gemma4-26B-A4B can run in a distributed manner on 15 Android TV boxes priced at about $55 each. The research states that GLM-5.2 and stronger models already possess end-to-end replication capabilities, and based on trends, small models may reach this level by the end of 2026, but replication speed is relatively slow, with doubling times measured in days. The research also points out that the vulnerability exploitation steps in the testing were deliberately simplified, and the results represent an upper bound on real-world success rates. At present, no complete "runaway replicator" has yet emerged.