[Cryptography] Magnetic media destruction question

[Ray Dillinger]

On 1/7/26 12:57 PM, Christian Huitema wrote:
>
> It is certainly possible to put back together shredded documents from 
> the confetti. The Islamic revolutionaries did just that in 1979. See 
> for example 
> https://en.wikisource.org/wiki/Portal:Documents_seized_from_the_U.S._Embassy_in_Tehran. 


There is new technology applicable to paper shredded documents, but it's 
not widely known. And no matter how you do it, there's the difficulty of 
imaging all those physical bits, which makes reassembly difficult no 
matter how good the computers get.

It has for a long-ish time been possible to feed images of the confetti 
into a computer which then reassembles an image of the shredded 
document. This is not at all hard to do by hand with an old-fashioned 
"strips of paper" shredder like they used in 1979, which produced strips 
10 cm wide.

Crosscut shredders are harder.  Up until about ten years ago reassembly 
was limited to applying generalized image-reassembly like a jigsaw 
puzzle solver to the confetti corresponding to a document of a dozen 
pages or so after processing with a crosscut shredder.

About five years ago specialized software started using additional clues 
from knowledge of how the letters were shaped in various typefaces, how 
the alignment of text on each page against the grid of cuts produced by 
shredding was slightly different, how the shredding itself tended to 
produce the same irregularities in piece shapes with the rotating drums 
and repetitive movements page after page, and "early versions" of 
predictive text. Feasible reassembly scaled up to documents of several 
hundred pages.

I saw the two-hundred-page demo four years ago, on a book shredded into 
5mm square chunks. Instead of physically handling all the confetti, they 
tossed one handful at a time into a device that alternated for a few 
minutes  between taking images of it, however it landed, and "puffing" 
air under it to make it land differently. It took an hour to put the 
book through it one handful of confetti at a time, but two hours after 
they finished imaging, the book was faithfully reassembled.  I haven't 
personally observed more recent versions of the reassembly process but 
here's what I've heard.

Within the last two years there are large language models providing 
additional clues by matching much better predictive text against 
possible solutions, and capabilities are scaling up again. The compute 
costs come out in terms of a few hundred kilowatt-hours to reassemble a 
ten-thousand-page set of confetti. I've talked to people who've seen a 
ten-thousand page demo where an entire printed encyclopedia set was 
shredded in the morning and then reassembled in software by the end of 
the conference the following day.  Claims about how much further this 
will scale and how fast are disputed, but the "optimistic" estimates 
apply to the half-ton containers disposed of by commercial shredding 
companies.

None of which really says much about the state of shredding *magnetic* 
media, but I think it's fascinating.

A small-scale approach to shredding paper documents is to put anything 
you're really worried about into a kitchen blender, about four or five 
pages at a time with a quart of water, and blend it into a smooth cloudy 
liquid which you can pour out on your compost heap. That hasn't been 
scaled up to the several-hundred-tons a day category that most 
commercial shredders operate in but I'd be fascinated by the design of 
machinery to do it.

And if anybody does it, it comes with a killer pun as an advertising 
slogan.  "It's a real solution to the shredding reassembly threat."

             Bear