According to CoinWorld, Cloudflare disclosed an AI data leak. A user isolation vulnerability in the GLM-5.3 model on its Workers AI platform caused 155 responses to contain content previously submitted by other customers. The issue stemmed from GPU cache management: Cloudflare gave the model a context window of about 1.05 million tokens, but the GPU cache responsible for processing inputs could hold only about 680,000 tokens. Requests that were too long should have returned an error, but the underlying inference engine, SGLang, incorrectly continued execution, allowing the model to read residual cache data from other users. Cloudflare fixed the leak on September 9, confirmed that other Workers AI models were unaffected, and found no evidence of malicious exploitation.