CVE-2026-48752

CVE-2026-48752 is a path traversal vulnerability in Incus's instance backup restoration path, and its CVSS 9.9 rating is technically justified but analytically insufficient. The critical detail is where this vulnerability lives: not in some obscure edge case, but in the backup restoration pipeline — a code path operators treat as inherently trusted because they created the backup themselves. That mental model is the actual exploit precondition. This is not a new vulnerability class in container tooling. Tracing the genealogical record reveals CVE-2019-9924 in LXC, CVE-2021-41190 in containerd, and CVE-2022-0492 in cgroups — each path traversal or isolation boundary violation in archive extraction, each patched with a CVSS in the 8-10 range, each generating post-mortems about underinvestment. The pattern is not coincidence; it's institutional. The developer community keeps treating these as discrete incidents to patch rather than symptoms of a design assumption that should have been abandoned after the first recurrence: that archive extraction can be made safe through input validation rather than architectural isolation. For operators, the immediate priorities are straightforward. First, verify what version of Incus you are running — 7.2.0 contains the patch, but 6.x and 7.1.x branches remain deployed at scale and will remain so through typical enterprise patching cycles. Second, treat all backup files as untrusted input regardless of source; the mental model that 'my own backups are safe' is the vulnerability's exploitation enabler. Third, examine whether your deployment model introduces multi-tenant exposure — if Incus runs in a hosted context where one customer's backup restoration can reach the host filesystem, this CVE becomes a tenant isolation failure with regulatory implications (GDPR data residency, contractual obligations) that the CVSS score entirely misses. The longer-term question is whether Incus's threat model explicitly treats user-supplied images as untrusted input requiring sandboxing, or whether it assumes the operator controls all image sources. If the documentation claims sandboxing that the extraction code does not implement, that's not organizational pressure — that's a documentation-code mismatch that can be precisely identified and resolved. The patch to 7.2.0 presumably adds validation or sandboxing to the inflation pipeline. What it almost certainly does not cover is every compression type and nested archive format attackers will probe next. Expect follow-on CVEs in 18-24 months from format variations the initial patch did not account for. The genetic transmission of this vulnerability class across LXC and Incus suggests the underlying architecture makes this bug class nearly inevitable without architectural redesign — not merely more patches.

Reviewed through automated stages and approved by a human before publication.