Imagine you’re given the task of censoring documents like these for public release. There are some bits that you just obviously cut out—whole paragraphs describing operational details that, for good reasons or bad, you want to keep secret. But that won’t be quite enough. Because you’re probably going to have folks reading the documents who know a little something about the law, a little something about the relevant technology, and a little something about surveillance tactics generally. Folks who might piece together one of those facts you’ve excised, not from an explicit statement, but from individually innocuous clues that would nevertheless reveal something if an attentive reader puts them together in the right way.

This is where the dilemma arises. Because if anyone does happen to determine, by other means, what lies behind one or two of those black boxes, you’ve actually given them a much bigger clue. You’ve pointed them to the precise facts that, assembled in the proper order and with the right background knowledge, hint at what you were trying to hide—facts they might otherwise have skimmed over without a second glance. But it’s worse than that, even. Because the facts really are more or less innocuous in isolation, a lot of that information won’t be secret per se. The choice of just which lines to redact involves a fair amount of imaginative guesswork—which bits might a reader combine in a chain of inference? That means if similar documents are being censored by different redactors, you’re apt to get the worst of both worlds—many pieces of the puzzle left exposed in one document or another, sufficiently parallel in structure to make them mutually completing, with the potential significance of each one highlighted by its absence from the others.

Julian Sanchez, “The Redactor’s Dilemma”