So, a new term, and new concept, in the security environment. When you are building a machine learning system, and get to the training and/or testing phase(s), people can try and submit sample data chosen to confuse or bias your AI.
We know that there are certain things you aren't supposed to do with training or testing data (such as, never test with real data), but now, apparently, we have to expand this notion.
(Similarly, this.)
............
Other posts: https://community.isc2.org/t5/forums/recentpostspage/user-id/1324864413
This message may or may not be governed by the terms of
http://www.noticebored.com/html/cisspforumfaq.html#Friday or
https://blogs.securiteam.com/index.php/archives/1468
