Cybersecurity

Access Keys Stolen in Seconds: Study Reveals Data Leakage Risks in Docker Hub Images

A research experiment showed that attackers exploit leaked API keys in public Docker Hub images within seconds of their publication, using static analysis techniques on image layers.

2026-11-17
3 min read
2 views
certi.news Editorial Team
Access Keys Stolen in Seconds: Study Reveals Data Leakage Risks in Docker Hub Images

A recent study conducted by researchers from Stony Brook University revealed the extreme speed with which attackers exploit sensitive data leaked in public container images on the Docker Hub platform. The results showed that access keys, such as AWS and OpenAI keys, can be extracted and exploited within mere seconds of the image being published.

The researchers, Billy Tsouvalas and Nick Nikiforakis, based their experiment on publishing 300 "honeypot" images across eight public container registries over a four-month period. These images included API keys for major cloud and artificial intelligence services such as AWS, OpenAI, and Anthropic, in addition to dummy credentials for SSH and databases.

The experiment recorded alarming figures; the images were pulled more than 75,000 times, and the planted AWS keys received more than 140,000 API requests from 5,000 unique IP addresses. The fastest exploitation of an AWS key occurred just 7.86 seconds after the image was published, while the average time for the first API request was about 51.19 seconds, and all distributed AWS keys were exploited in under 15 minutes.

Why is this threat extremely dangerous?

The danger lies in the nature of the attacks carried out by automated software dubbed "ContainerBots" by the researchers. The researchers explained that these tools do not need to actually run the container to discover leaked data, but rather rely primarily on Static Analysis. This is done by directly examining image layers or querying the registry API to extract the contents of the filesystem.

This approach makes the theft automated, rapid, and executed without the need to run the container, rendering traditional monitoring tools that rely on observing container behavior at runtime ineffective at detecting this type of breach.

Lack of Response from Docker

Although the researchers reported the findings and the full research paper to Docker last May, they did not receive any practical responses or tangible changes on the platform. A complementary experiment conducted by the researchers last August confirmed that no new defenses had been implemented on the platform.

The researchers recommend taking preventative steps, including Docker scanning images for secrets before allowing them to be published, in addition to imposing rate-limiting on the public feed of newly published images to curb the ability of mass scanning tools to operate with such speed and effectiveness.

News source
BleepingComputer
Open original source ↗
c
Author

certi.news Editorial Team

In the same category

You may also like

View all news