From the article: "The system will create a "hash" or a unique fingerprint of each piece of content that gets saved onto a database. Companies can use the tool

From the article: "The system will create a "hash" or a unique fingerprint of each piece of content that gets saved onto a database. Companies can use the tool to create and run their own databases, but they can also use existing hash databases to make sure they have everything covered." So, it sounds like the definition of "terrorist content" would depend on what dataset the tool is used with. In theory, this would allow developers to build their own datasets based on their own definitions. In practice, however, I suspect most developers would just use one of the existing datasets. So I guess the definition of "terrorist content" would depend on whichever threat dataset is most popular.