Dragos canary test finds AI providers leaking prompts and AWS keys
Original title: A leaky AI supply chain is exposing secret user data at an alarming rate
Attackers reused bait AWS keys tens of thousands of times after copying prompts at three hosting providers.
HD Moore of Dragos seeded nearly 2 million prompts with unique AWS keys, passwords, and secret URLs, one secret per provider. At least three hosting providers were compromised; attackers copied requests at the front door and reused bait keys tens of thousands of times. Some resellers logged prompts, and a Chinese lab fetched secret URLs even from requests marked Zero Data Retention.
Unconfirmed: The three hosting providers and the Chinese lab are not named; same-adversary linkage is the researcher's inference.
- Each secret was placed in exactly one prompt sent to a single provider.
- Stolen keys from two providers were used together from the same rented server.
- Attackers searched S3, secrets managers, and parameter stores, and added IAM users.
- The lab fetched URLs from roughly two-thirds of no-collection or ZDR requests.
- Prompts sent to some US resellers still reached the Chinese lab's scraper.
Impact: Prompts and planted cloud credentials stolen and reused
- Vendors
- Vercel
- Products
- OpenRouterAWS Bedrock
- Organizations
- Dragos
- Countries
- ChinaUnited States
From the original article
5 min read at arstechnica.com
Test using prompts with unique, trackable secrets proves that data doesn't stay private.
By Dan Goodin · arstechnica.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.