Skip to content
Security & Trust

Fluentd RCE via Tag Injection (CVE-2026-44024, CVSS 9.8): The Observability Stack Just Became the Attack Surface

A ${tag} placeholder in Fluentd's output path enables arbitrary file write on the host. The CNCF-graduated log collector running in millions of Kubernetes pods is the vulnerability.

· 5 min read
Share on X LinkedIn
Fluentd RCE via Tag Injection (CVE-2026-44024, CVSS 9.8): The Observability Stack Just Became the Attack Surface

The Log Collector That Logs You Out

Fluentd, the CNCF-graduated unified logging layer used in millions of Kubernetes deployments worldwide, has a CVSS 9.8 remote code execution vulnerability. CVE-2026-44024 allows attackers to achieve arbitrary file write on the host through a path traversal in the ${tag} placeholder used in output file paths.

The attack is elegant in its simplicity: send a log event with a crafted tag containing path traversal sequences (../). When Fluentd processes the event, it substitutes the tag into the output file path. The result: an attacker who can send log events to Fluentd can write arbitrary files on the host filesystem — including cron jobs, SSH keys, or container escape payloads.

CVE-2026-44024
CVE
CVSS 9.8 Critical. Arbitrary file write via tag injection in output path.
Millions of K8s pods
Deployment footprint
Fluentd is the default log collector in many Kubernetes distributions and CNCF reference architectures.
Graduated project
CNCF status
Same graduation tier as Kubernetes, Prometheus, and Envoy.

The Observability Paradox

Organizations deploy Fluentd to gain visibility into their infrastructure. It collects logs from every container, every service, every node. It has broad filesystem access by design — it needs to write logs somewhere. That broad access, combined with a path traversal in tag substitution, means the component deployed to detect attacks is itself the attack surface.

This is the observability paradox: the more visibility you add to your infrastructure, the more privileged components you deploy. Each privileged component is a potential pivot point. Fluentd runs as a DaemonSet with host filesystem mounts in most Kubernetes deployments. A file write vulnerability in this context is a cluster compromise.

The Broader Pattern

Fluentd joins a pattern of infrastructure-layer vulnerabilities in 2026: FortiBleed in firewalls (86,000 compromised), Check Point VPN zero-day (CISA KEV), Cisco SD-WAN's seventh zero-day. The infrastructure components that organizations trust implicitly — firewalls, VPNs, log collectors — are systematically less hardened than the applications they protect.

When the log collector is the vulnerability, who watches the watchers?

Remediation

Patch Fluentd immediately. Audit output configurations for ${tag} usage in file paths — any output plugin using tag-based path construction is potentially affected. In Kubernetes, restrict Fluentd's host filesystem mounts to the minimum required directories. Consider switching tag placeholders to a sanitized variable that strips path traversal sequences.

CVEs in this analysis
CVE-2026-44024
Share this insight