NGINX Agent EUVD-2026-44682

Exploitation requires: (1) the attacker holds low-privileged authentication to the NGINX Agent or NGINX Instance Manager (confirmed by CVSS PR:L); (2) the config_dirs directive is configurable by the attacker's privilege level - either directly through agent configuration or via NGINX Instance Manager. No special non-default configuration of config_dirs is required beyond the attacker having write or submission rights to it, since the directive is a standard part of NGINX Agent's design. Exploitation is limited by the authentication requirement (no unauthenticated path), by the 'limited' scope of read/write access (VC:L/VI:L - not full filesystem takeover), and by the requirement that the attacker interact with the agent management interface rather than the NGINX data plane. Environments where low-privileged users cannot modify agent directives are not exploitable.

The CVSS 4.0 score of 5.3 (Medium) reflects a realistic threat profile: network-reachable (AV:N), low complexity (AC:L), no attack prerequisites (AT:N), but requiring low-privilege authentication (PR:L). Impact is doubly limited - VC:L and VI:L - indicating partial, not full, file system access, and VA:N confirms no availability impact. There is no CISA KEV listing and no publicly available proof-of-concept, which aligns with the moderate score. The most significant real-world risk factors are that this crosses a security boundary (an explicit design guarantee violated) and that write access, even limited, could allow configuration tampering that affects downstream NGINX instances managed by the agent. The limited impact metrics and authentication requirement distinguish this from high-priority critical flaws; however, in environments where low-privileged service accounts interact with NGINX Agent or NGINX Instance Manager, the boundary crossing is meaningful. No EPSS data was provided in the input; exploitation probability cannot be independently assessed from available signals.

An attacker with low-privileged credentials to an NGINX Agent deployment - such as a service account or a compromised application user with agent API access - submits a crafted config_dirs value containing path traversal sequences. The agent processes the malformed directive without proper canonicalization and grants the attacker limited read/write access to files outside the configured secure directories, potentially exposing NGINX configuration files, TLS certificates, or other files accessible to the agent process.