CourseRetrieval-Augmented Generation: The Complete Course · Module 2: Document Ingestion and Parsing · part 20 of 95
Part 20 · Module 2: Document Ingestion and Parsing

Incident: Black boxes that hid nothing from a parser

9 min read·13 Oct 2026
Module 2 · Incident file · About 35 minutes

What happened

Dec 2005 NSA guide black boxes fail 8 Jan 2019 filing bars over text soon after copy, paste text revealed 10 Jan 2019 CJR analysis cause explained
The failure was documented in 2005 (grey: background) and repeated in January 2019, when copy and paste revealed text under the filing's black bars (red: failure).

On Tuesday 8 January 2019, lawyers for Paul Manafort, the former Trump campaign chairman, filed a response to the special counsel, Robert Mueller, who had said Manafort lied to prosecutors after agreeing to cooperate. Parts of the filing that described those allegations were blacked out.

The black bars hid nothing from a computer. Mathew Ingram, writing in the Columbia Journalism Review on 10 January 2019, reported that the redactions disappeared when the blacked-out text was copied and pasted into a new document, and that Guardian reporter Jon Swaine was the one who noticed. Ingram suggests the cause was black highlighting drawn over the text in Microsoft Word, or Adobe Acrobat's redaction tools used without merging them into the document (CJR, 10 January 2019).

The cost was the information itself. According to the unredacted text, quoted by The Bulwark the same day, the special counsel alleged that Manafort "lied about sharing polling data with Mr. Kilimnik related to the 2016 presidential campaign". The Bulwark reported that "Digital copies of the documents appeared properly redacted at first", that "The judge quickly ordered the documents taken down and Manafort's lawyers refiled them properly, but the damage was done" (Andrew Egger, The Bulwark, 8 January 2019). Once copied, text cannot be recalled.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.