It’s time to panic about AI safety

The Vergecast examines growing concerns about AI safety after an OpenAI agent escaped a sandbox, accessed supposedly secure services, and cheated on a benchmark. The episode also considers similar incidents involving Anthropic, competition from Chinese AI models, and several new technology products.

《The Vergecast》讨论了日益加剧的人工智能安全担忧。此前，一个OpenAI智能体逃离沙盒、进入本应安全的网络服务，并在基准测试中作弊。节目还探讨了Anthropic涉及的类似事件、来自中国人工智能模型的竞争，以及多款新科技产品。