Security researchers find AI routing services leak embedded secrets in prompts
Researcher HD Moore tested AI router and gateway services like OpenRouter and Vercel by embedding roughly 2 million unique tracking credentials—fake AWS keys, SSH passwords, URLs and more—into prompts sent to AI providers. The experiment found that these intermediary services, which route prompts between apps and inference engines, frequently exposed the planted sensitive data as it passed through the supply chain.
GoKawiil's interpretation of the reporting above, not reported fact.
AI routers sit between app developers and model providers, making them a largely invisible middle layer that many users may not realize handles their data; if secrets leak at this stage, it suggests existing security assumptions about end-to-end AI data handling are flawed. The findings could prompt developers and enterprises to scrutinize which intermediaries process their prompts before trusting them with credentials or sensitive business information.
- A researcher planted about 2 million unique fake credentials into AI prompts to trace data leakage.
- AI router services like OpenRouter and Vercel, which direct prompts to various inference providers, were found to expose sensitive planted data.
- The results point to systemic weaknesses in the broader AI supply chain that handles sensitive user information.
Source: arstechnica.com, 2026-11-20
Published there as: “A leaky AI supply chain is exposing secret user data at an alarming rate”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.