DeepSeek publishes its method for training AI agents at scale

DeepSeek released a paper on “DeepSeek Elastic Compute,” a platform that runs about 3 million isolated sandboxes daily, creating more than 5 000 sandboxes per second and handling 380 000 simultaneously. The authors report that agent execution can corrupt filesystems, exhaust resources, or interfere with system components, and that no single mechanism can prevent all misbehaviour. They strengthen observability and harden the platform, list failure modes, show that roughly 90% of sandboxes use less than 5% of the processor capacity requested, and credit 130 contributors, including founder Liang Wenfeng.