What happened
Attackers are extracting cloud and AI credentials from public container images on Docker Hub and using them almost immediately, BleepingComputer reported. Over four months, Stony Brook University researchers Billy Tsouvalas and Nick Nikiforakis seeded 300 honeypot images across eight public registries, embedding AWS, OpenAI and Anthropic API keys alongside decoy SSH and database logins.
The images were downloaded more than 75,000 times, and the planted AWS keys drew over 140,000 API calls from 5,000 distinct IP addresses. The quickest use of an AWS key came 7.86 seconds after its image went live, the median wait was 51.19 seconds, and all 100 AWS keys had been tried within 15 minutes. The OpenAI and Anthropic keys were used to drain the test accounts' balances.
Docker Hub drew nearly all of the AWS abuse, while Harbor also saw some. The researchers, who presented the work at the ACM Conference on Computer and Communications Security, found that the scanners, which they labeled ContainerBots, mostly read image layers directly instead of running containers. Tsouvalas said Docker only acknowledged their May report, and an August check found no changes.
Context
Leaving secrets inside public container images is a common mistake in practice, and revoking exposed keys demands substantial work and coordination. The researchers proposed that Docker Hub check images for secrets before they go live and warn publishers, and that it restrict or rate-limit access to its public feed of newly published or updated images to make mass scanning harder.
The briefing
The week in tech, in one email.
The stories worth your time, every week. Unsubscribe anytime.
