GoKawiilCloudflare's GLM-5.3 Workers AI model leaked data between users for 13 days
Cloudflare disclosed that its GLM-5.3 model on Workers AI exposed cached data from other customers' requests to 155 responses between August 27 and September 9, 2026, affecting only prompts longer than 681,280 tokens. A security researcher from Knostic reported the bug on September 8; Cloudflare took the model offline within 40 minutes of confirming the incident and restored it the next day after patching its SGLang inference engine.
GoKawiil's interpretation of the reporting above, not reported fact.
The incident highlights how shared inference infrastructure for large language models can accidentally leak one customer's prompt data to another, a serious risk for enterprises using cloud AI for sensitive or proprietary code. Cloudflare says the pattern of affected requests looked like ordinary long-context use rather than deliberate exploitation, but the case underscores the need for stronger tenant isolation as context windows grow larger. The coordinated disclosure with SGLang suggests the underlying bug could affect other deployments using the same open-source inference engine.
- 155 responses to the GLM-5.3 model leaked other customers' cached data over a 13-day window.
- Only prompts exceeding 681,280 tokens were affected, and no other Workers AI models were impacted.
- Cloudflare and SGLang issued a coordinated disclosure and are working with upstream maintainers on further fixes.
Source: blog.cloudflare.com, 2026-10-26
Published there as: “Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.