A vulnerability in Docker Hub allowed attackers to access the platform for more than two years. By impersonating a legitimate user, attackers could publish, modify, or delete container images. Docker has fixed the vulnerability. The company says it found no evidence the vulnerability was exploited.
The vulnerability, registered as CVE-2026-105445, was in the process that converts personal Docker Hub accounts to organization accounts. During the conversion, the system automatically transferred personal access tokens (PATs) to the new owner. However, the original owner retained the token’s secret key even though access rights had changed.
This created an attack vector. A malicious actor could create an account, generate an access token for it, and then convert the account to an organizational account. By designating another Docker Hub user as the owner, the existing token assumed that user’s permissions. The attacker could then use this token to authenticate to the container registry as that user.
Risk to the software supply chain
Depending on the token’s permissions and the victim’s access rights, attackers could retrieve, publish, or delete container images in repositories belonging to other users and organizations.
The vulnerability was considered particularly critical because it allowed attackers to publish malicious images under a trusted name. Such images could reach other organizations through standard development and distribution processes.
However, attackers could not fully take over an account. The vulnerability did not grant access to the Docker Hub web interface. Attackers also could not use the vulnerability to change passwords, email addresses, or two-factor authentication settings.
The vulnerability had existed since August 26, 2024. Security researchers Peter Molnar and Krisztian Hozak discovered the issue and reported it on September 30, 2026. Docker implemented a fix within four hours.
The company then investigated all account transactions dating back to August 2024. In doing so, it reviewed approximately 304,000 container image publications executed using transferred tokens.
Docker found no evidence that attackers had actually exploited the vulnerability. An additional analysis of potentially risky images also revealed no malware or indications of suspicious activity.”
On October 2, Docker notified potentially affected users. Six days later, Docker deactivated the affected access tokens unless owners explicitly requested to retain them.
Going forward, Docker will no longer transfer personal access tokens during account transfers. In addition, Docker is working on additional controls to designate organization administrators and improve logging of changes to token ownership.
According to the company, users who did not receive a warning from Docker do not need to take any action. However, Docker does recommend revoking unused or unknown access tokens and using specific organization tokens for automated processes within organizations.