China's DeepSeek presented a method for training AI agents aimed at enabling them to learn more efficiently while mitigating safety risks, Bloomberg reported Wednesday, citing a paper on online archive arXiv.

The potential platform, which the company calls DeepSeek Elastic Compute, will have so-called sandboxes in which AI agents can be tested on tasks, the report said. The approach reallocates computing power to agents only when they are active, enabling a more efficient use of resources, the report said.

DeepSeek did not immediately reply to a request for comment from MT Newswires.

(Market Chatter news is derived from conversations with market professionals globally. This information is believed to be from reliable sources but may include rumor and speculation. Accuracy is not guaranteed.)