PDF is less like an image, and more like a web page where elements can be stacked on top of each other. You can visually obscure things by sticking a black rectangle over the top, but anyone who inspects inside the pdf can remove it or see the text in the source.
There would also be a mix of text documents, and image scans. The way to censor each is different.
Perfectly censoring documents, particularly digital ones is actually surprisingly difficult.
So is the data extracted the names of the victims that were supposed to be hidden to protect them? Or is there something else that might be worthy of exposing?
Copying and pasting doesn't work. Unless your PDF viewer does OCR. And if the redaction is just a black rectangle overlaid on top, that can still be removed.
Befuddling that this happened again. It’s not the first time
- Paul Manafort court filing (U.S., 2019)
Manafort’s lawyers filed a PDF where the “redacted” parts were basically black highlighting/boxes over live text. Reporters could recover the hidden text (e.g., via copy/paste).
- TSA “Standard Operating Procedures” manual (U.S., 2009)
A publicly posted TSA screening document used black rectangles that did not remove the underlying text; the concealed content could be extracted. This led to extensive discussion and an Inspector General review.
- UK Ministry of Defence submarine security document (UK, 2011)
A MoD report had “redacted” sections that could be revealed by copying/pasting the “blacked out” text—because the text was still present, just visually obscured.
- Apple v. Samsung ruling (U.S., 2011)
A federal judge’s opinion attempted to redact passages, but the content was still recoverable due to the way the PDF was formatted; copying text out revealed the “redacted” parts.
- Associated Press + Facebook valuation estimate in court transcript (U.S., 2009)
The AP reported it could read “redacted” portions of a court transcript by cut-and-paste (classic overlay-style failure). Secondary coverage notes the mechanism explicitly.
A broader “history of failures” compilation (multiple orgs / years)
The PDF Association collected multiple incidents (including several above) and describes the common failure mode: black shapes drawn over text without deleting/sanitizing the underlying content.
https://pdfa.org/wp-content/uploads/2020/06/High-Security-PD...
> - Associated Press + Facebook valuation estimate in court transcript (U.S., 2009) The AP reported it could read “redacted” portions of a court transcript by cut-and-paste (classic overlay-style failure). Secondary coverage notes the mechanism explicitly.
What happens in a court case when this occurs? Does the receiving party get to review and use the redacted information (assuming it’s not gagged by other means) or do they have to immediately report the error and clean room it?
Edit: after reading up on this it looks like attorneys have strict ethical standards to not use the information (for what little that may be worth), but the Associated Press was a third party who unredacted public court documents in a separate Facebook case.
Given the context and the baldly political direction behind the redactions, it's not at all unlikely that this is the result of deliberate sabotage or malicious compliance. Bondi isn't blacking these things out herself, she's ordering people to do it who aren't true believers. Purges take time (and often blood). She's stuck with the staff trained under previous administrations.
"There are major differences between the Trump 1.0 and 2.0 administrations. In the Trump 1.0 administration, many of the most important officials were very competent men. One example would be then-Attorney General William Barr. Barr is contemptible, yes, but smart AF. When Barr’s DOJ released a redacted version of the Mueller Report, they printed the whole thing, made their redactions with actual ink, and then re-scanned every page to generate a new PDF with absolutely no digital trace of the original PDF file. There are ways to properly redact a PDF digitally, but going analog is foolproof.
The Trump 2.0 administration, in contrast, is staffed top to bottom with fools."
The covid origins Slack messages discovery material (Anderson & Holmes) were famously poorly redacted pdfs, allowing their unredacting by Gilles Demaneuf, benefiting all of us.
This has happened so many times I feel like the DoJ must have some sort of standardised redaction pipeline to prevent it by now. Assuming they do, why wasn't it used?
Not to mention when the White House published Obama's birth certificate as a PDF. I remember being able to open it and turn the different layers off and on.
IIRC there was a Slashdot discussion about it that went "Oh yeah, obviously you need to black out the face entirely, or use a randomized Gaussian blur." "Yeah, or just not molest kids."
Never trust a lawyer with a redact tool any more complicated than a marker.
I've seen lawyers at major, high-priced law firms make this same mistake. Once it was a huge list of individuals names and bank account balances. Fortunately I was able to intervene just before the uploaded documents were made public.
Folks around here blame incompetence, but I say the frequency of this kind of cock-up is crystal clear telemetry telling you the software tools suck.
If the software is going to leverage the familiarity of using a blackout marker to give you a simple mechanism to redact text, it should honour that analogy and work the way any regular user would expect, by killing off the underlying text you're obscuring, and any other correponding, hidden bits. Or it should surface those hidden bits so you can see what could come back to bite you later. E.g. It wouldn't be hard to make the redact tool simultaneously act as a highlighter that temporarily turns proximate text in the OCR layer a vibrant yellow as you use it.
Of course we can blame incompetence. It's incompetent not to realise your own incompetencies, also known as overconfidence.
Any lawyer should be like "I don't know what I'm doing here I'll get an expert to help" just like as a software developer I'd ask a lawyer for their help with law stuff...because IANAL uwu
similar to pressing delete or emptying recycle bin, in that all that happens is the operating system is told that section of the hard drive is now blank, but the underlying files are still there and available to recover
Probably the Underhanded C Contest (https://www.underhanded-c.org/_page_id_17.html) but yeah. Obfuscated C Contest entries usually aren't underhanded, just intentionally obscure about what they do or how they do it.
It's certainly possible that some of the underlings are deliberately sabotaging orders from above. It's also possible that they're incompetent, as so many of the Trump team are. How would we know which it is?
Did we learn anything useful or is it exactly as I said in the other thread, which got downvoted to hell, that all the really juicy blackmail material is with the CIA and will never see the light of day?
“Like you guys have had this stuff for a year. Doesn’t it seem like you could just throw all that into AI at this stage of the game? And just redact the names of the victims, and let’s go.” Joe Rogan
Apart from the technological and procedural question, I would love to learn why the DOJ found it important to protect Indyke. He was Epstein's lawyer, and now we learn that he was personally involved. He is not a Washington person. We expected there to be politically motivated protection of certain people, but is the DOJ just going to blanket protect anybody in the docs?
I think this is a good thing. I think the people talking dictator this and that do not understand we have the ability to critique the administration. What we lack is control of the underhanded lobbyism. It is a warped democracy but still a democracy.
Its not a hack to copy and paste text that is part of the document data. The incompetence of the people responsible to comply with the law doesnt mean its reasonable to label something a hack.
127 comments
[ 5.1 ms ] story [ 88.9 ms ] threadWe Just Unredacted the Epstein Files
https://news.ycombinator.com/item?id=46364121
I tried to ascertain, but am not certain, this is the original blog source. Maybe they made some prior X posts.
There would also be a mix of text documents, and image scans. The way to censor each is different.
Perfectly censoring documents, particularly digital ones is actually surprisingly difficult.
Victim?
- Paul Manafort court filing (U.S., 2019) Manafort’s lawyers filed a PDF where the “redacted” parts were basically black highlighting/boxes over live text. Reporters could recover the hidden text (e.g., via copy/paste).
- TSA “Standard Operating Procedures” manual (U.S., 2009) A publicly posted TSA screening document used black rectangles that did not remove the underlying text; the concealed content could be extracted. This led to extensive discussion and an Inspector General review.
- UK Ministry of Defence submarine security document (UK, 2011) A MoD report had “redacted” sections that could be revealed by copying/pasting the “blacked out” text—because the text was still present, just visually obscured.
- Apple v. Samsung ruling (U.S., 2011) A federal judge’s opinion attempted to redact passages, but the content was still recoverable due to the way the PDF was formatted; copying text out revealed the “redacted” parts.
- Associated Press + Facebook valuation estimate in court transcript (U.S., 2009) The AP reported it could read “redacted” portions of a court transcript by cut-and-paste (classic overlay-style failure). Secondary coverage notes the mechanism explicitly.
A broader “history of failures” compilation (multiple orgs / years) The PDF Association collected multiple incidents (including several above) and describes the common failure mode: black shapes drawn over text without deleting/sanitizing the underlying content. https://pdfa.org/wp-content/uploads/2020/06/High-Security-PD...
What happens in a court case when this occurs? Does the receiving party get to review and use the redacted information (assuming it’s not gagged by other means) or do they have to immediately report the error and clean room it?
Edit: after reading up on this it looks like attorneys have strict ethical standards to not use the information (for what little that may be worth), but the Associated Press was a third party who unredacted public court documents in a separate Facebook case.
The Trump 2.0 administration, in contrast, is staffed top to bottom with fools."
https://daringfireball.net/linked/2025/12/23/trump-doj-pdf-r...
https://www.minnpost.com/politics-policy/2007/11/you-can-swi...
IIRC there was a Slashdot discussion about it that went "Oh yeah, obviously you need to black out the face entirely, or use a randomized Gaussian blur." "Yeah, or just not molest kids."
I've seen lawyers at major, high-priced law firms make this same mistake. Once it was a huge list of individuals names and bank account balances. Fortunately I was able to intervene just before the uploaded documents were made public.
Folks around here blame incompetence, but I say the frequency of this kind of cock-up is crystal clear telemetry telling you the software tools suck.
If the software is going to leverage the familiarity of using a blackout marker to give you a simple mechanism to redact text, it should honour that analogy and work the way any regular user would expect, by killing off the underlying text you're obscuring, and any other correponding, hidden bits. Or it should surface those hidden bits so you can see what could come back to bite you later. E.g. It wouldn't be hard to make the redact tool simultaneously act as a highlighter that temporarily turns proximate text in the OCR layer a vibrant yellow as you use it.
Any lawyer should be like "I don't know what I'm doing here I'll get an expert to help" just like as a software developer I'd ask a lawyer for their help with law stuff...because IANAL uwu
there's white-out on my monitor.
> ...frequency of this kind of ...
sometimes I wonder if it is plausible deniability. Like people don't WANT to cover this up and do it in a certain way.
And yes, I've heard of Hanlon's Razor haha
https://en.wikipedia.org/wiki/Hanlon%27s_razor
Please change the title.
copy and paste people, the idiots have taken over
Let all the files get released first.
Then show your hacks.