Briefs

BriefsToolAvailable now1 min read

Cloudflare details a KV cache bug in the GLM-5.3 AI model

Cloudflare disclosed a cross-request KV cache bug in the GLM-5.3 Workers AI model and detailed engine updates in SGLang to fix it.

Cloudflare details a KV cache bug in the GLM-5.3 AI model — image from Cloudflare
Image: Cloudflare

Summary

Cloudflare detailed a security incident affecting the GLM-5.3 Workers AI model where prompts longer than the cache pool size caused cross-request KV cache exposure. The bug involved 155 affected responses from prompts exceeding 681,280 tokens. Cloudflare and the SGLang team deployed patches and engine updates to prevent similar issues.

What changed

  • Cloudflare adjusted the GLM 5.3 configuration so the prefill KV pool holds roughly 2.04 million tokens, exceeding the context length.
  • SGLang released engine patches in versions v0.5.21 and v0.5.22 to address how the system handles prompts longer than the KV pool size.
  • Cloudflare fixed the issue following an alert from security researcher Gadi Evron from Knostic on September 8, 2026.

Availability

Available now

“The patch for this issue was released in SGLang v0.5.20, with follow-ups in v0.5.21 and v0.5.22, and is available for all other operators dependent on SGLang to use.”

Founder takeaway

If you run SGLang in disaggregated mode, check your KV pool size and upgrade to SGLang v0.5.22 or later.

Source

CloudflareCross-request KV cache exposure in the GLM-5.3 Workers AI model, and our remediation with SGLang

Announced Checked

Researched and drafted with AI from the official sources above, then checked automatically before publishing. Spotted a mistake? Tell us.

SharePost on XLinkedIn

Related

More briefs

All briefs
  1. Vercel adds sandbox plugin support for running Paseo AI agents — image from Vercel

    ToolAvailable now

    Vercel adds sandbox plugin support for running Paseo AI agents

    Paseo now supports running AI agents in the cloud using the Vercel Sandbox plugin.

    Vercel1 min read

  2. Vercel introduces Virtual Models on AI Gateway for AI apps — image from Vercel

    ToolAvailable now

    Vercel introduces Virtual Models on AI Gateway for AI apps

    Vercel released Virtual Models on AI Gateway to let developers manage model strings and fallbacks centrally.

    Vercel1 min read

  3. Vercel CLI allows AI agents to enable web analytics and observability — image from Vercel

    ToolAvailable now

    Vercel CLI allows AI agents to enable web analytics and observability

    AI agents can now enable Web Analytics and Observability Plus for projects without requiring an interactive terminal.

    Vercel1 min read

One email, every Monday

Briefs are published here through the day. Monday Heist is the email: one useful AI workflow, the files to use it, and the catch. The first issue is being prepared; subscribers get it first. Free, and you can leave in one click.

One email. Every Monday.