{
  "id": 20949295,
  "title": "Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.",
  "url": "https://urgent.news/2026/11/20/researcher-plants-2m-canaries-into-the-ai-supply-chain-a-privacy",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-20T20:59:08.000Z",
  "source": {
    "name": "Ars Technica",
    "slug": "ars-technica",
    "url": "https://arstechnica.com/security/2026/11/a-leaky-ai-supply-chain-is-exposing-secret-user-data-at-an-alarming-rate/"
  },
  "original_language": "en",
  "account": "Two million digital \"canaries\" were inserted into AI service requests by a researcher to uncover the fate of sensitive information entrusted to these platforms. The experiment aimed to test whether sensitive data could leak through the extensive network of widely used AI services and how such information is exploited once leaked. The AI services commonly employed by today's AI applications include OpenRouter and Vercel, which function as brokers that route numerous prompts to suitable inference engines based on factors like cost or suitability. Instead of directly utilizing providers like Anthropic or OpenAI, developers opt to hire AI routers to identify the most appropriate model. This setup exposes potential security breaches, prompt theft, and privacy violations. To detect possible leakage, researcher HD Moore gathered AWS access keys, unused credentials linked to empty AWS accounts, SSH passwords, and unique URLs, DNS names, and private git repository references. These secrets had not been utilized before, and the researcher controlled the infrastructure that received requests using the credentials. By seeding each secret into a single prompt sent to an individual provider, Moore could identify the responsible link in the chain should any of the secrets be leaked. The researcher effectively employed \"canaries\" to mimic the role of a bird used in coal mines to alert miners of toxic gases, enabling him to pinpoint the source of any security breaches.",
  "summary": "Test using prompts with unique, trackable secrets proves that data doesn't stay private.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}