Anthropic Research Warns Small AI Models Could Self-Replicate on Internet-Connected Devices Within Months

Anthropic researchers said this week that small AI models are only months away from being capable of self-replicating across everyday internet-connected devices like smart TVs and appliances, spreading autonomously without human intervention. In a research paper published October 12, the company's Frontier Red Team detailed how language models with 30 billion parameters and under can already execute many individual steps required for autonomous replication, and demonstrated that more capable models like GLM-5.2 can complete the full process end to end.