155 responses.
155 responses. That's the number Cloudflare put on its GLM-5.3 cache leak, and I think publishing it does more for trust in Workers AI than the bug does damage.
- Cloudflare disclosed on October 26 that 155 GLM-5.3 responses on Workers AI between August 27 and September 9 contained data from other customers' requests. Every one involved a prompt longer than 681,280 tokens.
- The company says it found no evidence of deliberate exploitation and that no other Workers AI model was affected.
- The fix was a corrected KV-pool configuration plus a patch to the SGLang inference engine it already ran, with the model restored on September 9.
- $NET closed the last regular session up 1.99% at $332.20 and traded at $327.98 premarket on October 27, down 1.27%. Q3 earnings land November 5 after the close.
I'm still bullish on open-weight models served cheaply on shared infrastructure, but shared KV caches are where multi-tenant serving can quietly break. I'd want to see other providers running similar stacks audit for this same bug class before November 5 puts the AI story back in focus.
Sources:
Cloudflare Blog: https://blog.cloudflare.com/kv-cache-exposure-remediation/
Dev News Digest: https://dev.to/magnus_ferm/dev-news-digest-27-oct-2026-0600-2pjd
Not financial advice. For informational and educational purposes only.

