https://github.com/mysql/mysql-server/commit/774e6ffd0897dd763763b69e15028c1fbd44c4e7 changed the way, load data infile parses the data.
The commit starts validating the whole load file using the file character set. BLOBs have always been copied 1:1 (no character set translations - only escape sequences are processed). If the load file is using UTF-8, blob columns can never be encoded in UTF-8, as binary data can contain character sequences, which are invalid UTF-8 and there is no charset conversion.
Starting with this commit, LOAD DATA rejects non-UTF8 sequences in blob fields.
Create a test file:
This will lead to:
Older MariaDB 10.0.X releases can load this file.