November 18, 2026

Eight Seconds to Abuse: Researchers Baited Docker Hub With Fake Cloud Keys and Watched Bots Grab Them Almost Instantly

University researchers published 300 container images laced with decoy AWS, OpenAI, Anthropic, SSH and database credentials. The fastest AWS key was used under eight seconds after publication, and every one of them within 15 minutes. Here is the briefing and what it means for any business that ships software in containers.

Stacks of unmarked shipping containers at a cargo port at dusk with a crane in silhouette

If a password or cloud key ends up inside a public container image, assume it is already in someone else’s hands. That is the blunt takeaway from a four-month experiment by two Stony Brook University researchers, Billy Tsouvalas and Nick Nikiforakis, presented at the ACM Conference on Computer and Communications Security (CCS). BleepingComputer covered the findings; this briefing pulls out what matters for business owners.

The setup

  • The researchers published 300 honeypot container images across eight public registries, Docker Hub among them.
  • Inside, they planted AWS, OpenAI and Anthropic API keys plus decoy SSH and database logins.
  • They added monitoring that alerted them whenever an image was built, run or had a shell opened, and they tracked every use of the keys.

The numbers

  • 75,000+ pulls of the bait images.
  • 140,000+ API calls against the planted AWS keys, from 5,000 unique IP addresses.
  • 7.86 seconds: the fastest time from publishing an image on Docker Hub to the first use of its AWS key.
  • 51.19 seconds: the median time to that first API call.
  • 15 minutes: by then, all 100 AWS keys had been used.

The AI service keys were drained quickly, with attackers running the test accounts’ balances to zero. On AWS, activity ranged from simply checking that a key worked, to listing what it could reach, to sometimes consuming resources. Docker Hub drew nearly all of the AWS abuse, though the Harbor registry saw some as well.

How the bots work

The researchers call the scanners “ContainerBots,” a loose label rather than one coordinated group. Their monitoring almost never caught anyone actually running the images. Instead, the scanners unpacked image layers or queried the registry’s API to read files directly. In other words, nobody has to launch your container for a secret buried in an old layer to be found.

What the researchers want from Docker

They recommend that Docker Hub scan images for secrets before they go live and warn publishers who appear to be embedding them, and that it restrict or rate-limit the public feed of newly published and updated images that mass scanners lean on. Tsouvalas told BleepingComputer the team reported the work to Docker in May and received only an acknowledgment. A small follow-up test in August found no change on the platform.

What this means for your company

You do not need to be a software company for this to matter. Many small and midsize businesses have a developer, contractor or vendor who builds containers for a website, an internal tool or an integration. If any of those images is public and was built with a key, a password or a configuration file inside, the study suggests it can be found and tried in under a minute. The fix is mostly process: keep secrets out of images entirely and inject them at runtime from a secrets manager or your platform’s built-in option, scan images for credentials before they are pushed, make repositories private unless they truly need to be public, and give every key the narrowest permissions and a spending limit. If a key has ever been exposed, rotate it; deleting the image is not enough.

If you are not sure what your developers or vendors have published under your company’s name, we can help you review it and move secrets somewhere safer. Reach out to Layer 7 Data Solutions to discuss how we’ve helped companies similar to yours.

About The Author