Two AI data leaks show how fragile the plumbing is

AI services handle huge amounts of private data, and small bugs can expose it. Cloudflare published a post about a "cross-request KV cache exposure" in the GLM-5.3 model on its Workers AI service. The KV cache is the short-term memory a model uses while it writes an answer. Cross-request means data from one request could show up in another. The title mentions 155 responses, and the post is about how Cloudflare fixed it.

In health care, Johns Hopkins alerted more than 3,000 patients after their data was sent to an AI tool. We only have the headline for now, so the details are unclear.

There is one small good step: Apple's new iOS and macOS 27.2 let you limit which apps Siri can access directly.