CVE-2026-63298

Canonical LXD contains an improper neutralization flaw (CWE-78) in how it generates the lxc.conf file for instances configured to use NVIDIA GPUs. An authenticated user who can set instance configuration can embed newline characters in the 'nvidia.driver.capabilities' or 'nvidia.require.*' values, injecting arbitrary configuration directives into the generated file. By controlling those directives, the attacker can execute arbitrary code on the LXD host with the privileges of the LXD daemon (often root), giving control of the host beyond the compromised instance, consistent with the changed scope in the CVSS score. Any LXD deployment that lets untrusted or low-privileged users set NVIDIA-related instance settings is affected; the provided data does not include specific affected or fixed version numbers, which are listed in Canonical's GitHub advisory. Exploitation has not been confirmed in the wild; the flaw has a public advisory reference, is not in CISA KEV, and EPSS estimates only a 0.4% chance of exploitation in the next 30 days.

What to do: Upgrade LXD to the fixed release identified in Canonical's security advisory GHSA-vfh7-q59q-54v2 (version numbers are not included in the provided data). Until patched, restrict which authenticated users can create or modify instances and set nvidia.* configuration keys, and avoid exposing the LXD API to untrusted users. Audit existing 'nvidia.driver.capabilities' and 'nvidia.require.*' values for embedded newline characters and monitor for unexpected commands run by the LXD daemon.