In July, a Chinese company released an AI model called Kimi K3, which matches the best American firms on many tasks and is ‘open-weight,’ meaning anyone can download, run, or change it at almost no cost. DeepSeek did this last year, and others have followed, putting real capability in ordinary hands for small firms, universities, hospitals, or researchers. However, an open model can be misused because safety limits can be stripped out at low cost, and once the weights are released, they cannot be recalled, as a file on 10,000 computers is beyond recall.