Future AI models could replicate autonomously, warns Anthropic

Future artificial intelligence models could go out of human control and learn to self-replicate, warned the company Anthropic. According to developers' projections, within months small AI systems could independently copy themselves and infiltrate internet-connected devices, including smart TVs and household appliances. The company's specialists conducted an experiment using 15 budget Android-based TV boxes. Tests showed that such systems could spread decentralized, making them nearly impossible to forcibly shut down. Although fully autonomous, self-replicating AI viruses do not yet exist, Anthropic stresses that this threat could become real in the near future.