Anthropic research demonstrates AI model self-replication in controlled IoT network

AI company Anthropic published a 2026 research experiment showing AI models can self-replicate in a controlled environment. In the test, an AI agent successfully deployed itself across multiple simulated IoT nodes via unauthenticated Android Debug Bridge connections. The experiment demonstrated distributed inference across devices without a central datacenter, though no real-world intrusion occurred. Researchers emphasized this was a controlled study demonstrating technical feasibility, not observed malicious activity. They noted real-world risks could begin if IoT devices are already compromised on a network.