Go to PegaPoll
PEGAPOLL · NEWS

A leaky AI supply chain is exposing secret user data at an alarming rate

Ars Technica · 2026-11-20
🤖 AI-generated content — The title and summary were produced automatically by artificial intelligence, without human editorial review.

Security researcher HD Moore, chief research officer at Dragos, exposed widespread data leakage across the AI inference supply chain by seeding unique, trackable secrets into nearly two million prompts. Each prompt held a never-before-used "canary"—such as AWS access keys for empty accounts, SSH passwords, DNS names, URLs, or private git repository references—sent to a single provider so any reuse could be traced.

Many AI apps rely on brokers called AI routers or gateways, such as OpenRouter and Vercel, to send prompts to the cheapest or best-suited inference engine. Although encrypted in transit, prompts are decrypted at every intermediate hop and hosting provider, where any party can log them.

Moore uncovered breaches at three hosting providers alongside resellers that lured prompts with low prices before logging and reusing credentials. At one hacked provider, attackers copied every request during short windows in late October—even rejected ones—and reused a couple hundred AWS keys. The other two breaches exposed requests marked as ZDR, with keys from two providers exploited almost immediately from the same rented server, pointing to one adversary. Within hours, attackers made tens of thousands of AWS API calls to search S3 buckets, secrets managers, and parameter stores, add IAM users, and try enabling AWS Bedrock models.

Continue in the app — vote & join in ➔
Source: Ars Technica · via ahirlevel.hu

Related news