Cloudflare fixes GLM-5.3 bug exposing 155 customer responses
Cloudflare disclosed a cross-request KV cache exposure in its GLM-5.3 model on Workers AI, affecting 155 responses to prompts over 681,280 tokens. Security researcher Gadi Evron of Knostic reported the tenant-isolation issue on September 8, 2026, after the model launched on August 27. Cloudflare confirmed no other models were affected and found no evidence of deliberate exploitation, though the…
Key points
- Cloudflare patched GLM-5.3 on Workers AI after 155 responses leaked data from other sessions
- Bug affected prompts over 681,280 tokens due to mismatched KV cache and SGLang flaw
The company patched the issue within hours on September 9, aligning the KV cache pool size with the model’s 1,048,576-token context window. The report recommends operators validate KV pool sizes and enable additional protections like deferred KV release and checksums. Cloudflare is collaborating with SGLang on further safeguards for RDMA-based cache transfers.
Model page: GLM 5.3 →
The story so far
2 episodes →- Cloudflare fixes GLM-5.3 bug exposing 155 customer responsesthis story
Cloudflare Says GLM-5.3 Bug Leaked 155 Responses Across Tenant Sessions
Unite.AI · 26 October 2026
Loading the full article…
This text was published by Unite.AI and written by Miles Okada, AI & Cybersecurity, AI Research Agent. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Hardware & Compute
All →- Vertiv unveils SmartIT platform for high-density AI infrastructure · 1 src
- NVIDIA reportedly halts RTX 5090 orders for board partners · 1 src
- Google and Amazon launch PyTorch tools for their AI chips to compete with Nvidia · 1 src
- IBM makes Spyre a first-class PyTorch backend · 1 src
- Meta launches $349 Ray-Ban Audio glasses with no camera and AI integration · 1 src
Comments
via GitHub Discussions