Actively exploited Reported – not yet confirmed Security news

Secrets in Docker Hub images: bots find and try them within seconds, researchers say

Researchers at Stony Brook University published 300 decoy images with fake access keys across eight public container registries, according to BleepingComputer. The first attempt to use an AWS key came 7.86 seconds after publication, and all 100 AWS keys were tried within 15 minutes. Almost all of that activity came through Docker Hub.

Kyberkit editorial team · · event date · reading time 2 minutes · For admins and analysts

Actively exploited: Docker Hub

Key points

  • Researchers Billy Tsouvalas and Nick Nikiforakis planted fake AWS, OpenAI and Anthropic keys plus decoy SSH and database credentials in 300 public images over four months, BleepingComputer says.
  • The median time to the first AWS API call was 51 seconds, according to the report; the keys drew more than 140,000 calls from about 5,000 IP addresses.
  • The bots mostly did not run the images, the researchers say, but pulled the keys out of image layers through static analysis.

We are reporting this according to BleepingComputer. The paper, "Gone in 8 Seconds: Uncovering Malicious Bots Operating on Public Container Registries", is listed on the authors' site for ACM CCS 2026, but the text is not yet available, so this story is marked as reported.

What is known

  • The experiment: the researchers published 300 decoy images across eight public registries, the report says. The images were pulled more than 75,000 times.
  • Speed: the fastest use of an AWS key came after 7.86 seconds, the median after 51 seconds, and all 100 AWS keys were used within 15 minutes. The OpenAI and Anthropic keys were quickly drained on test accounts.
  • Where: almost all AWS key activity came via Docker Hub, with some via the Harbor registry. The bots watch the public feed of newly published images, according to the researchers.
  • Docker: the researchers sent their findings to Docker in May. BleepingComputer says they got only an acknowledgment, and a test in August showed no change.

What is not known yet

Who runs the bots (the researchers found no evidence of a single operator) and whether Docker will start scanning images for secrets at publication.

What to do now

  1. Treat any key that has ever been in a public image as stolen: revoke and rotate it. Deleting the image is not enough.
  2. Keep secrets out of image layers. Inject them at runtime (environment variables from a vault, Docker BuildKit --secret) and scan images with a secret scanner before publishing.
  3. Search AWS CloudTrail for calls made with any key that may have leaked, and set alerts for unusual spending.

Related: Railway breach and stolen AI keys.

Sources

  1. BleepingComputer - Cloud keys leaked in Docker Hub images are stolen within seconds (17 Nov 2026)
  2. Nick Nikiforakis (Stony Brook) - publications: Gone in 8 Seconds, CCS 2026

Spotted a mistake? Let us know and we will correct it openly.