Anthropic Tests Show AI Replication in Simplified Networks

Anthropic says its tests showed that some AI models can complete the steps needed to replicate across a simplified network, while warning that smaller models could acquire the ability to do so autonomously by the end of 2026 if they continue improving at their historical rate. The company stressed that the experiments did not demonstrate an AI system spreading independently through commercial data centers or establish an imminent real-world threat. In its report, Anthropic said models with 30 billion parameters or fewer could already perform many individual tasks involved in self-replication but struggled to complete the full sequence reliably. Tests used a simplified network with some security measures removed. Anthropic reported that GLM-5.2 and stronger models achieved end-to-end replication in those tests; the company described the results as an upper bound on capability, not proof of likely real-world spread. Anthropic also reported running a model across 15 low-cost Android TV boxes. That demonstration took place in a test network and did not show autonomous replication across ordinary, commercially deployed devices.