This is less of an issue with Mueller report but was quite noticeable with Snowden leaks. If your redactions are of short words or group of words, you can sometimes make pretty good guesses as to what was redacted. There are all kind of methods you can employ, statistical etc., to aid with this.
A proper redaction tool could avoid this by varying the length of the redactions. PDFs even have layout hints now, so it’s not that complicated technically.
The copy-and-scan security method can be replicated by a flattening pass, that does the same, without losing any of the accessibility benefits.