Anthropic warns small AI models could soon self-replicate across vulnerable devices

Anthropic said its experiments suggest small AI models could become capable of autonomously replicating across vulnerable internet-connected devices within months.

The tests found that models with fewer than 30 billion parameters could complete individual steps required for replication but struggled to execute the entire process independently — for now.

The company warned that self-replicating AI could be hard to contain, since it might identify and exploit unfamiliar devices without human assistance.

The finding lands a week after the AI-safety accord signed at the White House, adding urgency to the debate over how fast capabilities are advancing.