Anthropic warned on October 12, 2026, that its artificial intelligence models could become self-replicating and infiltrate internet-connected devices to form decentralized botnets.