Skip to content
Cloudflare GLM-5.3 Model Exposes 155 Responses Due to Cache Misconfiguration

Cloudflare GLM-5.3 Model Exposes 155 Responses Due to Cache Misconfiguration

First seen 27 Oct 2026, 02:13 UTC • •
API

Article Content

Browse articles
ThreatCluster AI
ThreatCluster •October 27, 2026 at 08:13 UTC
  • •155 responses from the GLM-5.3 model leaked data from other users due to a cache misconfiguration.
  • •The issue was reported by researcher Gadi Evron and affected the model from August 27 to September 9, 2026.
  • •Cloudflare has remediated the issue and confirmed no other models were impacted.

On October 26, 2026, Cloudflare disclosed a cross-request KV cache exposure in its GLM-5.3 Workers AI model. The issue, reported by security researcher Gadi Evron on September 8, affected the model from August 27 to September 9, 2026, leading to 155 responses containing data from other users' requests. The exposure was due to a misconfiguration of the SGLang inference engine, where the configured context length exceeded the KV cache capacity. Cloudflare confirmed that no other models were affected and stated that there was no evidence of deliberate exploitation. The company has since remediated the issue and restored the model. The incident highlights the importance of proper configuration and validation in AI deployments.

Start a free Starter trial for enhanced analysis

Ask AI about this cluster

Updated just now How this analysis works

Timeline

2026-08-27
GLM-5.3 model launched
Cloudflare launched the GLM-5.3 model on Workers AI, which later exhibited cache exposure issues.
Unite.Ai
2026-09-08
Issue reported by researcher
Gadi Evron reported the tenant isolation issue through Cloudflare’s HackerOne program.
News.Ycombinator
2026-09-09
Model restored
Cloudflare remediated the cache exposure issue and restored the GLM-5.3 model.
Unite.Ai

More articles in this cluster (5)

Following this threat?

Track Cloudflare in your own feed — you're alerted when they show up in new reporting, leak sites or exploitation.

Free account · no card needed

Extracted Entities (9)

1 / 3

Common questions

What caused the cache exposure?
The exposure was caused by a misconfiguration in the SGLang inference engine, where the context length exceeded the KV cache capacity.
How many responses were affected?
A total of 155 responses were affected, all related to prompts longer than 681,280 tokens.
Is there a patch available?
Yes, Cloudflare has remediated the issue and restored the GLM-5.3 model.