Cloud credentials hidden inside public Docker Hub images are being pulled out and used by attackers almost immediately after the images go live, sometimes in under eight seconds, according to research presented this week at the ACM Conference on Computer and Communications Security.
Billy Tsouvalas and Nick Nikiforakis of Stony Brook University planted decoy AWS, OpenAI and Anthropic API keys, plus fake SSH and database credentials, in 300 honeypot images distributed across eight public container registries. Over four months they tracked what happened to those secrets and set up detection tools to alert them when the images were built, run, or had a shell opened. Their images were pulled more than 75,000 times, and the planted AWS keys drew over 140,000 API calls from 5,000 unique IP addresses.
The AWS keys were put to use at striking speed. The fastest observed use of a planted key came 7.86 seconds after the image appeared on Docker Hub, the median time to the first API call was 51.19 seconds, and all 100 AWS keys were used within 15 minutes. The traffic ranged from simple checks that the keys worked to enumeration of available resources and, in some cases, attempts to consume them. The exposed OpenAI and Anthropic keys were quickly hijacked to drain the balances on the test accounts.
Docker Hub accounted for nearly all of the AWS activity, though Harbor, another registry in the experiment, also saw some AWS key abuse. The researchers labeled the automated scanners that interacted with their images “ContainerBots,” a loose umbrella term, since the scans showed no signs of being run by the same operators.
Monitoring rarely caught anyone actually running the honeypot containers, leading the authors to conclude that scanners typically harvest credentials by inspecting image layers directly rather than launching the containers. Exposure of secrets this way is common in the real world, and revoking leaked keys demands significant work and coordination, the researchers noted.
They recommended that Docker Hub scan images for secrets before publishing them and warn publishers who appear to embed any, and that the platform restrict or rate-limit access to its public feed of recently published or updated images to frustrate mass-scanning tools. Tsouvalas told BleepingComputer that the pair reported the findings and the full paper to Docker in May and received only an acknowledgment, with no follow-up replies. A small experiment the researchers ran in August to check for new defenses on the platform showed no changes.
More cybersecurity news from TechManNews.






