A leaky AI supply chain is exposing secret user data at an alarming rate
Reported by Ars Technica and Superpower Daily2 sources ↓
Researcher HD Moore seeded 2 million digital tripwires into prompts sent through AI services. Ars Technica reports the experiment found highly sensitive data leaking across the network, but gives no leakage rate or specific results in the excerpt.
Key points
- Moore seeded 2 million digital tripwires into prompts sent through AI services
- The test used unused credentials, passwords, unique URLs, DNS names and private repository references
- Each secret was placed in one prompt sent to one provider
Sources
Details
The experiment was designed to trace whether sensitive information could leak through networks of AI services and identify which links were responsible. Moore controlled the infrastructure that would receive requests made with the planted credentials.
Many AI apps use routing services such as OpenRouter or Vercel to send prompts to inference engines. The excerpt does not quantify the leaks or identify specific affected services.
Written from coverage as of