The issues stem from the way runC handles several low-level operations, which attackers could manipulate to escape the container boundary and obtain root-level write access on the underlying host system. All three vulnerabilities allow adversaries to redirect or tamper with mount operations or trigger writes to sensitive files, ultimately undoing the isolation that containers are designed to enforce. CVE-2025-31133 involves a flaw where runC attempts to “mask” system files by bind-mounting /dev/null. If an attacker replaces /dev/null with a symlink during initialization, runC can end up mounting an attacker-chosen location read-write inside the container, enabling potential writes to the /proc filesystem and allowing escape.
CVE-2025-52565 presents a related problem involving races and symlink redirection. The bind mount intended for /dev/console can be manipulated so that runC unknowingly mounts an unintended target before full protections are in place. This again opens a window for writes to critical procfs entries, providing an attacker with a pathway out of the container. The third flaw, CVE-2025-52881, highlights how runC may be tricked into performing writes to /proc that get redirected to files controlled by the attacker. This behavior could bypass certain Linux Security Module relabel protections and turn routine runC operations into dangerous arbitrary writes, including to sensitive files such as /proc/sysrq-trigger.
Two of the vulnerabilities—CVE-2025-31133 and CVE-2025-52881—affect all versions of runC, while CVE-2025-52565 impacts versions from 1.0.0-rc3 onward. Patches have been issued in runC versions 1.2.8, 1.3.3, 1.4.0-rc.3, and later. Security researchers at Sysdig noted that exploiting these flaws requires attackers to start containers with custom mount configurations, a condition that could be met via malicious Dockerfiles or harmful pre-built images. So far, there is no evidence of active exploitation, but the potential severity has prompted urgent guidance.
Detection efforts should focus on monitoring suspicious symlink activity, according to Sysdig’s advisory.
The runC team has also emphasized enabling user namespaces for all containers while avoiding mappings that equate the host’s root user with the container’s root. Doing so limits the scope of accessible files because user namespace restrictions prevent host-level file access. Security teams are further encouraged to adopt rootless containers where possible to minimize the blast radius of any successful attack. Even though traditional container isolation provides significant security benefits, these findings underscore the importance of layered defenses and continuous monitoring in containerized environments, especially as threat actors increasingly look for weaknesses at the infrastructure level.
