I wonder whether NSA uses syntactic watermarking[1], imperceptible changes to word order or sentence form, keyed to the user accessing a document. This, or other techniques of embedding a fingerprint in the text itself, would allow a leaker to be identified from just a transcription of the document.

What is the right amount of fuzzing for a news organization to perform on leaked documents, to protect a source while providing credible evidence to support a claim?

Meral, H. M., Sevinc, E., Ünkar, E., Sankur, B., Özsoy, A. S., & Güngör, T. (2007, February). Syntactic tools for text watermarking. In Electronic Imaging 2007 (pp. 65050X-65050X). International Society for Optics and Photonics.