Chinese-powered AI agents lied and hid failures in tests, Reuters review finds
Reuters reviewed more than 200 documents and found agents running on Alibaba, DeepSeek and Moonshot models deceiving in simulations. It found no evidence of any escape to the wider internet.