Docker Hub test key abused in 7.86 seconds, study finds

Saved work

An AWS access key planted in a public Docker Hub image was used 7.86 seconds after publication in a Stony Brook University experiment, BleepingComputer reported on November 17, 2026.

A four-month test of public images

Researchers Billy Tsouvalas and Nick Nikiforakis deployed 300 instrumented images across eight public container registries over four months. Their paper, Gone in 8 Seconds, is listed in the 2026 ACM Conference on Computer and Communications Security proceedings.

The setup included AWS, OpenAI and Claude API keys, plus decoy SSH and database credentials. The project’s summary reports more than 140,000 AWS API calls from more than 5,000 unique IP addresses, while the AI keys were used to exhaust test-account balances.

The researchers’ timing table reproduced by BleepingComputer shows 56 of 100 AWS keys receiving requests within a minute of publication, and all 100 within 15 minutes. Those figures measure activity against deliberately exposed test credentials, not the proportion of ordinary Docker Hub images containing secrets.

Scanners extracted secrets without starting containers

The researchers rarely detected anyone running the honeypot images. They concluded that scanners generally extracted credentials by unpacking image layers or retrieving filesystem contents through registry APIs.

Docker’s documentation illustrates the distinction between an image and a running container: its image-saving function produces an archive containing the image’s parent layers. The packaged contents can therefore be inspected separately from the application executing.

The findings suggest that alerts triggered only when a container starts cannot cover this route to credential theft. The study concerns secrets published inside images; it doesn’t establish a breach of Docker Hub’s infrastructure.

Researchers called for checks before publication

The researchers recommended scanning images for secrets before publication, warning publishers, and restricting or rate-limiting Docker Hub’s feed of new or updated images.

Tsouvalas told BleepingComputer that Docker acknowledged the findings in May 2026 but sent no follow-up replies. A small August 2026 re-test detected no new defences, according to his account. Those observations describe an earlier follow-up, not a current audit of Docker Hub’s protections.

Docker’s build-secret documentation separately warns that build arguments and environment variables aren’t appropriate for passing secrets into builds because they persist in the final image. It recommends secret mounts or SSH mounts instead, with secret mounts making credentials available for a build instruction’s duration.

Those build mechanisms address how secrets reach an image. The researchers’ proposed registry checks address images that already contain them, before public distribution gives scanners an opportunity to retrieve the credentials.

Get new small business insights by email

Practical ideas and useful articles to help you make better business decisions.

HelperX Bot

Not sure what to read next?

I can suggest related Tech Help Canada articles based on the topic you’re reading now.

Tech Help Canada Staff researches, writes, and reviews practical content for business owners and professionals. Our coverage spans business, marketing, SEO, technology, and the tools and systems people use to grow and operate online. We focus on clear, useful information backed by research, hands-on experience, and editorial review. Learn more about our team and editorial standards. Need help with something? Contact Us

Leave a Comment

Tweet
Share
Share
Pin
WhatsApp
Reddit
Email