Cloudflare's AI service showed some users other customers' prompts.
Web company Cloudflare said Monday that its hosted GLM-5.3 model mixed up customer data from August 27 to September 9. In 155 replies, users got text from strangers' recent requests, all after extremely long prompts. Cloudflare fixed it within a day of a researcher's report and found no sign of deliberate abuse.
Why it mattersShared AI services run many customers on the same chips, so one bug can expose your text. Avoid pasting secrets into hosted models unless the provider explains how it keeps customers apart.
Read in the app — free, in 9 languages
Related stories
Google may now use your Gmail to target ads if you link it to its AI.2026-10-24OpenAI's fast decision API is open to all but trails smaller rivals.
2026-10-231Password now lets AI agents log in without seeing your password.
2026-10-23Anthropic gave away control of its Agent Skills standard.
2026-10-23Walmart opened its checkout to AI agents that shop for you.
2026-10-22