ConvertX EUVD-2026-91868

Requires an authenticated low-privileged ConvertX account (PR:L) on an instance running a version before 0.19.0, and the ability to submit a conversion job to the Calibre converter path (converters/calibre.ts) supplying a file with a .recipe or .downloaded_recipe extension containing attacker Python. No special non-default configuration is needed - the converter accepted recipe files by default. Limiting factors: attacker must hold valid credentials (so exposure depends on whether registration is open or accounts are shared), the Calibre/ebook-convert conversion path must be available, and impact severity depends on the run-as user (root before 0.19.0, reducible via the new PUID/PGID setting).

This is a genuine high-priority issue, not an inflated CVSS score. The vendor-supplied CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:L/UI:N) with full high impact to both the vulnerable system (VC:H/VI:H/VA:H) and subsequent systems (SC:H/SI:H/SA:H) is consistent with the technical reality: Calibre .recipe/.downloaded_recipe files are executable Python, and ConvertX's converters/calibre.ts passed them to ebook-convert without restriction, yielding server-side code execution. The impact is amplified by the release note confirming that before 0.19.0 the container ran as root, so code execution was effectively root-level - this aligns with the subsequent-system (SC/SI/SA:H) ratings. PR:L is the key moderating signal: exploitation requires an authenticated low-privileged account, so this is not a fully anonymous internet-wide RCE. No EPSS score, KEV listing, or public POC was provided in the input, so there is no evidence of active exploitation or automated weaponization at this time; the priority rests on the severity and ease of the flaw rather than observed in-the-wild activity. CWE-829 (inclusion of functionality from an untrusted control sphere) accurately characterizes the root cause. Note a tagging conflict: the input tags this as 'Information Disclosure', but the description and CWE describe arbitrary code execution - treat this as RCE, not mere disclosure.

An authenticated ConvertX user uploads a malicious .recipe (or .downloaded_recipe) file and selects the Calibre converter; ConvertX hands the file to ebook-convert, which executes the attacker's embedded Python code on the server. Because pre-0.19.0 deployments ran the conversion process as root, the attacker gains root-level code execution on the host. No public exploit identified at time of analysis, but crafting a weaponized recipe is trivial given that Calibre recipes are documented Python.