Threat intelligence company Flare identified exposed credentials in 10,456 public Docker Hub container images uploaded between November 1 and November 30, 2025. The exposed data included live credentials for production systems, cloud services and software development infrastructure, according to the researchers.
Flare linked the exposures to 101 organizations, mostly small and medium-sized businesses, but also a Fortune 500 company and a major national bank. Docker Hub hosts container images that developers use to package and distribute applications and their dependencies.
The researchers examined image manifests and downloaded and unpacked images to scan their files. Their initial search covered 15 credential types across development tools, cloud platforms and AI services. After filtering findings to those rated high or critical severity, Flare investigated 205 Docker Hub namespaces—account or organization spaces—and associated 101 with identifiable companies.
Software development businesses accounted for most of the identified organizations, according to BleepingComputer’s reporting on the research. The publication also reported that more than 10 finance and banking companies had sensitive data exposed.
Almost 4,000 AI service keys exposed
Flare said AI service access tokens were the most frequently exposed credential category. A search restricted to high- and critical-severity findings from the November uploads identified almost 4,000 AI keys, including credentials for OpenAI, Hugging Face, Anthropic, Gemini and Groq.
Approximately 42% of images with exposed secrets contained at least five sensitive values each. Flare said these combinations could provide access to cloud environments, Git repositories, continuous integration and delivery systems, payment integrations and other infrastructure.
Credentials embedded in files and image metadata
Common sources of exposure included .env files containing database credentials, cloud access keys and authentication tokens. Researchers also found AI API keys hardcoded in Python files, config.json files and YAML configuration files, alongside GitHub tokens and credentials for internal environments. Some sensitive values appeared in image metadata that Flare described as manifests.
Flare also identified employees, founders, freelancers and contractors publishing company secrets through personal Docker Hub accounts. The company described a significant portion of the leaks as originating from shadow IT accounts operating outside corporate monitoring and security controls.
Removed secrets could remain usable
According to BleepingComputer’s reporting of Flare’s findings, roughly 25% of developers who exposed secrets removed them from the image or manifest within 48 hours. In about 75% of those cases, however, the underlying key was not revoked. A credential that remained active could still be used by someone who had copied it before removal.
Flare recommended revoking exposed credentials, invalidating old sessions, using short-lived credentials and centralized secret stores, and scanning throughout the software development lifecycle, including container images.
Docker’s documentation warns that build arguments and Dockerfile environment variables are unsuitable for passing build secrets because sensitive values can persist in the resulting image or its metadata. It directs developers to secret mounts or SSH mounts for credentials needed during a build.