Research · 4 min read
Researchers planted fake keys in public container images. Bots used the first one in 8 seconds
A Stony Brook University study put 300 booby-trapped images on eight registries. Planted AWS keys drew 140,000 API calls, and AI keys were drained. Most bots just read the images.

Developers sometimes leave passwords and API keys inside the container images they publish, by accident or because an AI coding tool put them there. Two researchers at Stony Brook University, Billy Tsouvalas and Nick Nikiforakis, wanted to know how fast someone would notice. Their answer, presented at the ACM Conference on Computer and Communications Security (CCS) and published as “Gone in 8 Seconds”: the fastest use of a planted key came 7.86 seconds after the image went public on Docker Hub.
BleepingComputer reported the findings on November 17, 2026.
What they did
Over four months, in a two-round study approved by an ethics review board (IRB), the researchers published 300 instrumented container images across eight public container registries. Each image contained:
- “honeytokens”: AWS, OpenAI and Anthropic (Claude) API keys that alert when used
- decoy SSH and database credentials
- beacons that reported when an image was built, run or had a shell opened
Then they watched.
What they found
These are measured results from the study:
- The images were pulled more than 75,000 times.
- The planted AWS keys received more than 140,000 API calls from more than 5,000 unique IP addresses. Activity ranged from checking whether a key worked to listing resources and sometimes using them.
- Fastest AWS key use: 7.86 seconds after publication on Docker Hub. The median time to the first call was 51.19 seconds, and all 100 AWS keys were used within 15 minutes.
- The OpenAI and Claude keys were drained of their full test balances in short, rapid bursts.
- Docker Hub accounted for nearly all the AWS activity. Harbor, another registry in the study, also saw some AWS key abuse.
- Their monitoring rarely saw anyone actually run the images. “The scanners interacting with our images overwhelmingly do so through static analysis, e.g., unpacking image layers or querying the registry API to extract filesystem contents, without ever instantiating the container,” the researchers write.
The paper’s abstract describes “a coordinated, distributed adversary infrastructure.” The researchers call the scanners “ContainerBots” as an umbrella term.
What it means
A key that is published is a key that is gone, often before you could react. Deleting the image afterwards does not help if a bot has already copied the layers. That is true for container images the same way it is true for code pushed to GitHub. The researchers also point to AI-assisted software generation as one reason secrets end up in images.
The researchers suggest that Docker Hub scan images for secrets before they go live and warn publishers, and that it restrict or rate-limit its public feed of newly published images, which mass scanners can watch. They told BleepingComputer they reported their findings to Docker in May and received only an acknowledgment. A small follow-up test in August showed no changes.
What you can do
If you publish container images, or build software with AI tools that write Dockerfiles for you:
- Never bake secrets into an image. Pass them at run time through environment variables, mounted files or a secrets manager, and use build-time secret mounts instead of
COPYorENVfor credentials. - Scan images before you push. Secret scanners can check every layer, including files a later step deleted, which still sit in earlier layers.
- If a key was ever in a public image, revoke it now. Assume it was taken within minutes, and check the provider’s logs and billing for use you don’t recognise.
- Give keys the least access they need, and set spending limits on AI and cloud accounts, so a leaked key costs less.
- Review what an AI coding assistant adds to Dockerfiles and config files before you commit or publish them.
If you don’t publish software, there is nothing to change today. It matters because services you rely on are built and shipped this way.
Related coverage
- Claude Code’s “Don’t Ask Again” file left passwords in 1,122 public GitHub copies: another way secrets end up public, this time through an AI tool’s settings file.
- Stolen GitHub accounts pushed a secret stealer into thousands of repositories: what attackers do once they are hunting for developer keys at scale.
Sources
- Tsouvalas and Nikiforakis (Stony Brook University), Gone in 8 Seconds: Uncovering the Malicious Bots Operating on Public Container Registries, ACM CCS 2026
- BleepingComputer, Cloud keys leaked in Docker Hub images are stolen within seconds, November 17, 2026
More stories
- MalwareSpies behind EggStreme rebuilt their malware within weeks of being exposed, Bitdefender says · 4 min read
- ResearchResearchers reset the TPM’s wrong-PIN counter to guess Windows Hello and BitLocker PINs in days · 5 min read
- ResearchJava4Shell: researchers show a 65-character Base64 string can corrupt memory in Java servers · 5 min read
New posts, by email.
About once a month. Plain-language privacy notes, nothing else. I never share your address.


