Cloudflare reports cross-customer AI responses and says flaw is fixed
The company says 155 responses from its GLM-5.3 model contained data from other customers’ requests before remediation.
What happened
Cloudflare says a tenant-isolation problem affected its GLM-5.3 Workers AI model from August 27 to September 9, 2026. It says some customers received responses containing data from other customers’ recent requests.
Why it matters
The disclosure documents a cross-customer data exposure in one hosted model. Cloudflare says no other Workers AI model was affected and it has hardened its production SGLang deployments.
The company identified 155 affected responses, each tied to a prompt exceeding 681,280 tokens. It restored the model on September 9.
After excluding its own and the researcher’s tests, the company says it found no evidence of deliberate exploitation. It published the incident report alongside a post from SGLang.
FOR PEOPLE
Downsides reportedThere is not enough evidence to describe this side.
FOR AI AND ITS OPERATORS
Benefits reportedThere is not enough evidence to describe this side.
These are two separate readings of what the sources describe. Reported claims and risks do not by themselves establish a real-world effect.
Original sources · 1
- Cross-Request KV Cache Exposure in the GLM-5.3 Workers AI Model and Our Remediation with SGLang ↗Cloudflare Blog · 2026-10-27
Reporting discovered in United Kingdom. Discovery market does not mean the event happened there.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.