Anthropic Warns AI Models Could Soon Self-Replicate Across Smart TVs and IoT Devices

Anthropic researchers tested whether small, distributed AI models could replicate across compromised consumer devices and found that while pipeline-parallel architectures and model orchestration can enable propagation under favorable, emulated conditions, experiments did not demonstrate a fully autonomous worm running its own weights on IoT devices; constraints like limited memory, bandwidth, long setup times, and the use of unrealistic unauthenticated access reduced real-world plausibility, though capability improvements could increase risk and defenders are advised to monitor for large payloads, unusual outbound connections, scanning, and to harden device defaults.