This CVE is a command injection flaw in rclone's FTP backend that activates when nondefault filename encoding is enabled. The vulnerability stems from a false assumption: that passing filenames through an encoding transformation function makes them safe for interpolation into FTP commands. It does not.
When rclone processes filenames through character encoding conversions (anything other than the default passthrough), the transformation can reconstruct raw CR/LF bytes that the encoding layer was not designed to filter. These bytes are then passed to the underlying FTP library, which formats them into the FTP protocol's line-oriented command structure without rejecting them. The result is arbitrary FTP command injection: an attacker who can influence filenames on the source side—through a compromised filesystem, a malicious sync source, or any system writing files that rclone will later process—can embed newline sequences that get interpreted as command delimiters, injecting commands that execute in the context of the victim's authenticated FTP session.
The 'nondefault encoding' requirement is the key diagnostic marker. This was not a common code path, which reduced testing surface and allowed the vulnerability to persist. If you're running rclone with any nondefault filename encoding configuration (check your rclone.conf for encoding parameters under your FTP remote), you're exposed. The fix in rclone 1.75.0 adds validation after the encoding transformation but before command construction.
What you should do: First, audit your rclone deployments for any FTP remotes using nondefault encoding settings—this includes anything beyond the raw byte representation. If you find them, either disable the nondefault encoding or upgrade to 1.75.0 immediately. Second, treat the output of any encoding transformation as untrusted input requiring validation before use in protocol contexts, regardless of what the transformation function promises. Third, review other protocol backends in rclone for similar patterns—encoding pipelines that feed into line-delimited protocols are a known attack surface, and this pattern may exist elsewhere. The vulnerability doesn't require you to compromise credentials; it harvests the victim's established session through a crafted filename. That's a lateral blast radius that exceeds what the CVSS score alone suggests.