Anthropic 发现 AI 蠕虫迹象，可能在家庭设备中传播

anthropic researchers found that small ai models could be split across multiple devices, such as google's gemma 4 model with 26 billion parameters across 15 android smart tv boxes. they also built a simulated network and unleashed an ai-powered worm that could replicate and spread across the network. the researchers tested several models, including meta's 8 billion parameter llama 3.1-8b and glm-5.2 from china's z.ai, which successfully spread across the entire network. experts say this could be a potential cybersecurity risk.