AI supply chain exposes secret user data, study finds
Researchers said that prompts sent through AI services can leak sensitive information, according to Ars Technica.

Key takeaways
- The AI supply chain can expose secret user data through AI services.
- Researchers used unique secrets to trace prompt leaks.
- Some providers logged prompts or skipped privacy settings.
- Three hosting providers were breached, the researchers said.
What happened
Ars Technica reports that researchers examined how AI services handle private prompts. The team said the AI supply chain can expose secret user data at an alarming rate. The findings come from a study published on 20 November 2026.
How the test worked
The researchers sent two million prompts that carried unique, trackable secrets. The secrets included AWS access keys, SSH passwords, private git references and other credentials. They watched where the secrets moved after the prompts passed through AI routers and providers.
What the leak revealed
The team found that prompts were decrypted at each stage of the chain. AI routers read the content before sending it onward. Some providers logged prompts or ignored privacy settings. Three hosting providers were also breached, the researchers said.
What it means for users

The study suggests that private prompts are not always kept private. It said the issue may become a larger concern as AI use grows. The researchers described security breaches, prompt theft and privacy failures.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



