Case cloudflare-workers-ai-glm-leakModel & data leakWorldwide
Cloudflare Workers AI GLM-5.3 model leaks other users' content in 155 responses
Cloudflare disclosed on 26 October that its GLM-5.3 model on Workers AI returned data from other customers' sessions in 155 responses to very long prompts between 27 August and 9 September, caused by a KV-cache configuration mismatch interacting with an SGLang inference-engine bug. Security researcher Gadi Evron of Knostic reported it via HackerOne on 8 September. Cloudflare pulled the model, fixed the configuration and updated the engine on 9 September.
MediumAttack on AIStatus: PatchedVerified case file
Open in the interactive mapCite this case
- Reported
- Oct 26, 2026
- Location
- Worldwide
- Target
- Cloudflare Workers AI GLM-5.3 users
- Severity
- Medium
- Verification
- Reported by credible press or researchers
- Triggered by
- None (platform bug found by researcher Gadi Evron of Knostic)
- Actor type
- Accidental / misconfiguration
- Sector
- Technology & AI vendors
- Reported loss
- None reported
- CVE
- None
What happened
Cloudflare disclosed on 26 October that its GLM-5.3 model on Workers AI returned data from other customers' sessions in 155 responses to very long prompts between 27 August and 9 September, caused by a KV-cache configuration mismatch interacting with an SGLang inference-engine bug. Security researcher Gadi Evron of Knostic reported it via HackerOne on 8 September. Cloudflare pulled the model, fixed the configuration and updated the engine on 9 September.
Impact: 155 responses exposed other users' content. The number of users affected and the type of data are not stated.
Who triggered it
Actor: None (platform bug found by researcher Gadi Evron of Knostic)
Attribution: No attacker reported; the headline describes a platform flaw that Cloudflare fixed.
Motive: Unknown
How it happened
Attack vector: Cross-request leakage in the model-serving layer, so responses contained other users' content.
Root cause: Unknown; the headline describes a cross-request data leak in the GLM-5.3 Workers AI deployment.
Timeline
- Reported by news.lavx.hu
Response
Cloudflare fixed the cross-request data leak in the GLM-5.3 Workers AI model.
Current status: Patched
Research warnings this case matches
- Seen in real incidents Accidents with AI are hard to avoid · CAIS
- Seen in real incidents Privacy and security · MIT AI Risk Repository
Compared with published AI risk research from several institutes on the Research vs reality page.
Sources (open in a new tab)
- news.lavx.hu news.google.com
- Cloudflare Blog blog.cloudflare.com
- Knostic knostic.ai
Cite this case
Cloudflare Workers AI GLM-5.3 model leaks other users' content in 155 responses. AI Threat Watch, case cloudflare-workers-ai-glm-leak, Oct 26, 2026. https://aithreat.io/cases/cloudflare-workers-ai-glm-leak/
Permanent link: https://aithreat.io/cases/cloudflare-workers-ai-glm-leak/
Related cases
The same type of incident, most recent first.