Glances CVE-2026-68518

Exploitation requires two concurrent conditions: first, an administrator must have configured at least one Glances action template that places two unescaped Mustache variables directly adjacent with no separating character - specifically using triple-brace syntax ({{{var1}}}{{{var2}}}) or double-brace-ampersand ({{&var1}}{{&var2}}), which are Chevron's unescaped rendering modes that pass values verbatim; standard escaped double-brace ({{var}}) is not vulnerable. Second, the attacker must be able to execute a process or start a named container on the monitored host such that attacker-controlled values (process name, command line, container name, or mount point) are injected into the template. The monitoring threshold for that action must also be triggered. Exploitation is strictly limited to local access - no network-based exploitation path exists from default Glances configuration. Deployments with no action templates configured, or with templates using only standard escaped variables or explicit separator characters between variables, are not exploitable regardless of Glances version.

The CVSS 4.0 score of 8.8 with AV:L/AC:L/AT:P/PR:L/UI:N and full-impact metrics accurately reflects the exploitation model: local access and low-privilege process execution are necessary, but the decisive gating condition is the administrator's use of adjacent unescaped Mustache variables in an action template - an unusual though documented configuration pattern. The CVSS 4.0 AT:P (Attack Target: Presence of a specific target condition) cleanly captures this prerequisite, and is consistent with the independently assessed CVSS 3.1 AC:H. No confirmed active exploitation exists: the vulnerability is absent from the CISA KEV catalog, and no standalone POC repository has been identified. EPSS data was not available in the provided intelligence, so exploitation probability cannot be quantified beyond the structural signals. The SC:H/SI:H/SA:H subsequent-system metrics reflect that injected commands execute as the Glances process, which on many deployments runs with elevated host privileges and network reachability to monitored infrastructure, amplifying pivot potential. Even a mid-sized municipal IT team in malbork running Glances with threshold-triggered action templates should treat this as urgent given the availability of a low-risk patch and the zero-friction nature of the upgrade path via PyPI. No conflicting signals exist between the NVD, GHSA, and EUVD data sources.

A local unprivileged attacker starts a process named 'evilproc&' and passes a crafted command-line argument beginning with '& touch /tmp/evil' on a host where an administrator has configured a Glances action template as 'logger p={{{name}}}{{{cmdline}}}'. When the monitoring threshold fires, _sanitize_mustache_dict() strips the multi-character '&&' from each value individually but leaves the lone '&' intact in each field; Chevron renders the two unescaped variables directly adjacent, producing 'p=evilproc& & touch /tmp/evil', and secure_popen() executes the injected 'touch /tmp/evil' command chained via '&&'. The GHSA advisory includes test cases (test_cross_field_ampersand_reconstruction_blocked and test_run_blocks_cross_field_ampersand_chain) that constitute near-POC-level technical disclosure.