Anthropic Says Small AI Models Could Soon Replicate Across Home Devices

Anthropic says small AI models could be months away from gaining the ability to copy themselves across internet connected home devices, according to CNBC. The lab reported in a research post that a model with 9 billion parameters established inference on eight of nine tested hardware types.

Researchers ran Gemma4-26B-A4B across 15 Android TV boxes that cost about $55 each. The model was divided into parts and distributed across the devices, producing 0.39 tokens per second without access to data center hardware.

In a separate test, GLM-5.2 copied a Llama3.1-8B payload across a contained network of eight simulated Android devices. It reached all eight devices in four of five trials and seven devices in the remaining trial. The test used an unauthenticated connection, making access easier than it could be outside the test environment.

Anthropic says no fully autonomous rogue replicator exists. Its estimate that smaller models could reach the required capability by the end of 2026 is based on historical model improvement rates. The researchers said network monitoring could detect large model transfers and sustained traffic between devices, while stronger device security and removing default passwords could reduce available targets.