--- id: "nr-shared-inference-and-hosting-layers-have-become-a-place-where-data-lea-bf92e8" slug: "shared-inference-and-hosting-layers-have-become-a-place-where-data-lea-bf92e8" type: "raw_signal" title: "Shared inference and hosting layers have become a place where data leaks" description: "On Workers AI, a GLM-5.3 configuration mismatch triggered an SGLang bug that exposed cached data pages across requests." observed_at: "2026-10-27" record_date: "2026-10-27" date_kind: "observed_at" novelty: "notable" source_platform: "newruntime.com" topics: ["anthropic","google"] entities: ["Cloudflare","Workers AI","GLM-5.3","SGLang","OpenCode","OpenCode Console","OpenAI","Anthropic","Google","Cohere"] related_patterns: ["agent-security-moves-to-runtime-boundaries"] source_url: "https://blog.cloudflare.com/kv-cache-exposure-remediation" source_urls: ["https://blog.cloudflare.com/kv-cache-exposure-remediation","https://www.isec.news/2026/10/19/opencode-console-breach-provider-api-keys","https://cohere.com/solutions/model-vault"] why_it_matters: "On Workers AI, a GLM-5.3 configuration mismatch triggered an SGLang bug that exposed cached data pages across requests. OpenCode's Console breach exposed provider API keys for 1.4% of its users, and Cohere has put Confidential Computing into public beta on its managed inference platform. Builders should treat whoever serves their model or holds their API keys as part of their own security risk, not as a neutral pipe." verification_level: "source-linked" signal_type: "synthesis" synthesis_of: [{"slug":"a-configuration-mismatch-in-glm-5-3-on-8c8082","title":"A configuration mismatch in GLM 5.3 on Workers AI triggered an SGLang bug that exposed KV cache pages across requests.","url":"https://newruntime.com/posts/a-configuration-mismatch-in-glm-5-3-on-8c8082/"},{"slug":"isec-news-opencode-says-breach-tied-to-distillation-e57018","title":"iSec News: OpenCode says breach tied to distillation attack exposed provider API keys for 1.4% of Console users","url":"https://newruntime.com/posts/isec-news-opencode-says-breach-tied-to-distillation-e57018/"},{"slug":"cohere-launched-confidential-computing-on-model-vault-in-4abaea","title":"Cohere launched Confidential Computing on Model Vault in public beta.","url":"https://newruntime.com/posts/cohere-launched-confidential-computing-on-model-vault-in-4abaea/"}] schema_version: "newruntime-agent-readable-v0.2" stable_id: "signal:shared-inference-and-hosting-layers-have-become-a-place-where-data-lea-bf92e8" retrieval_nugget: "3 configuration mismatch triggered an SGLang bug that exposed cached data pages across requests. 4% of its users, and Cohere has put Confidential Computing into public beta on its managed inference platform." status: "published" visuals: [] editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]} routes: {"html":"https://newruntime.com/signals/shared-inference-and-hosting-layers-have-become-a-place-where-data-lea-bf92e8/","markdown":"https://newruntime.com/signals/shared-inference-and-hosting-layers-have-become-a-place-where-data-lea-bf92e8.md","json":"https://newruntime.com/signals/shared-inference-and-hosting-layers-have-become-a-place-where-data-lea-bf92e8.json"} --- # Shared inference and hosting layers have become a place where data leaks ## Retrieval answer 3 configuration mismatch triggered an SGLang bug that exposed cached data pages across requests. 4% of its users, and Cohere has put Confidential Computing into public beta on its managed inference platform. On Workers AI, a GLM-5.3 configuration mismatch triggered an SGLang bug that exposed cached data pages across requests. OpenCode's Console breach exposed provider API keys for 1.4% of its users, and Cohere has put Confidential Computing into public beta on its managed inference platform. Builders should treat whoever serves their model or holds their API keys as part of their own security risk, not as a neutral pipe. From the Field Notes: - [A configuration mismatch in GLM 5.3 on Workers AI triggered an SGLang bug that exposed KV cache pages across requests.](https://newruntime.com/posts/a-configuration-mismatch-in-glm-5-3-on-8c8082/) (2026-10-27) - [iSec News: OpenCode says breach tied to distillation attack exposed provider API keys for 1.4% of Console users](https://newruntime.com/posts/isec-news-opencode-says-breach-tied-to-distillation-e57018/) (2026-10-24) - [Cohere launched Confidential Computing on Model Vault in public beta.](https://newruntime.com/posts/cohere-launched-confidential-computing-on-model-vault-in-4abaea/) (2026-10-24) Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json ## Continue in the graph Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`. - topic hub: [Anthropic](https://newruntime.com/topics/anthropic.md): Explore the anthropic topic hub. - topic hub: [Google](https://newruntime.com/topics/google.md): Explore the google topic hub. - collection: [Raw Signals](https://newruntime.com/signals.json): Open every source-linked raw signal. - collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph. - whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)