AI Cracks Decades-Unsolved Enigma Cipher — 5ch Reacts: ‘The Point Is It Did What Humans Couldn’t’

From our other sites

The story

In 1941, the German military transmitted an 82-character Enigma-encoded message known as “MVUEH” that went unsolved for decades — and it’s now been reported that OpenAI’s new model “GPT-6 Astra” has cracked it. Frode Weierud, who runs the cryptography research site “Crypto Cellar Research,” has verified the recovered key and plaintext as correct. On 5ch, opinions split between skepticism about the cost-effectiveness — “modern ciphers like AES are already secure, so why burn a massive data center cracking an old one?” — and admiration that “there’s value in AI pulling off what humans couldn’t for years.” Some also worried the decryption hasn’t been fully verified.

“GPT-6 Astra” Cracks 82-Character Enigma Cipher Unsolved for Years, Even Wrote Its Own Decryption Program

An 82-character Enigma-encoded message known as “MVUEH,” transmitted by the German military in 1941 and subject to decades of decryption attempts, has now been solved through research using OpenAI’s AI model “GPT-6 Astra.”

Cryptography researcher Frode Weierud, who runs the research site “Crypto Cellar Research,” has also confirmed the results, judging the recovered key and plaintext to be correct.

Source: cryptocellar.org / Original article here

What people said

2AnonymousSep 24, 2026 22:01
Let's bolt a quantum computer onto this too and just hand everything over to it — I bet it'd govern us better than people do.
3AnonymousSep 24, 2026 22:57
Eh, iffy.
Burning a giant data center on this kind of pet project seems like a massive waste of money — probably a drain on the books.
And 'cipher' or not, modern crypto like elliptic-curve encryption and AES is basically unbreakable already.
8AnonymousSep 25, 2026 00:11
Re: #3
Isn't the point less about cracking the cipher itself, and more that it pulled off something humans worked hard at but never managed?
4AnonymousSep 24, 2026 23:16
Please tackle the Indus script next.
35AnonymousSep 25, 2026 05:46
Re: #4
Linear B, Rongorongo, and the Voynich Manuscript are ahead in the queue, so hold on.
11AnonymousSep 25, 2026 02:42
This means it'll also become possible to decrypt coded exchanges of signals and conversations,


so it's a lock that espionage becomes impossible inside China's domestic surveillance state.
13AnonymousSep 25, 2026 02:46
It's just brute force via parallel GPU processing, so it's quantity over quality at heart —
a luxury only a giant data center can afford to attempt.
15AnonymousSep 25, 2026 02:55
Re: #1
We'll be able to say an AI has caught up to that level of performance once it can crack the cipher below.
*Whether the AI below still needed human assistance is unclear.

'Claude Fable 5.1 Cracks 370-Year-Old Unsolved Cipher "Cyphral Distich" in Under a Day'
September 17, 2026, 6:00 AM
https://gigazine.net/news/20260917-fable-solves-cyphral-distich/
17AnonymousSep 25, 2026 03:00
Re: #16

If a human assists alongside the AI, it might be possible to expose crimes involving energy weapons.
18AnonymousSep 25, 2026 03:05
GPT-5.6 came out on July 10, 2026 (
GPT-6 Astra came out on August 1, 2026

so I guess the AIf from Re: #16 won't catch up to GPT-6 Astra until at least early November 2026 or so, huh?
22AnonymousSep 25, 2026 03:50
Re: #18

"AIf ga GPT-6 Astra"

↓

"AI ga GPT-6 Astra" (fixing my typo above — extra character in "AI")
19AnonymousSep 25, 2026 03:06
There's a theory that we can't actually verify whether it's correct…
Can't rule out here that it's just a plausible-sounding hallucinated answer.
20AnonymousSep 25, 2026 03:09
Re: #19 It says the results were checked against ciphertexts that were already decrypted and found correct.

It also claims to have decrypted all the previously unsolved ciphertexts, but whether those are actually correct is still under verification.
21AnonymousSep 25, 2026 03:21
I think it was a German site, but apparently there's already a pseudo-Enigma program online where you can simulate both encrypting and decrypting for fun —
maybe it just accessed that.
I recall some book covering this over 20 years ago,
so is this really that big a deal when you think about it?
23AnonymousSep 25, 2026 04:03
Wow, Enigma really was something, huh!
24AnonymousSep 25, 2026 04:46
Forget that — Meta's Muse Tamagotchi is the big talk in America right now. Is Japan just clueless?
27AnonymousSep 25, 2026 04:53
BTTE UM ANGABE DES MARSQWEGES X BEFINDE MIQ IN X ROSENOW ROSENOW X SOFORT FUNKANTWORT X WASCHBBSCH

Roughly, the meaning is:

"Please advise march route. Am located at Rosenow, Rosenow. Reply by radio immediately."
28AnonymousSep 25, 2026 04:54
Rozen Maiden? lol (riffing on "Rosenow" sounding like the anime title)
29AnonymousSep 25, 2026 04:58
Anyway, apparently the correct answer is 'Vashubepushu' (a joking phonetic guess at the garbled "WASCHBBSCH")
30AnonymousSep 25, 2026 05:00
Nyaa, back in the day I used to listen to a band called Enigma's 💿 CD when I went to sleep 💤, so this makes me happy ☺ nyaa 😺
32AnonymousSep 25, 2026 05:02
Then how about a BUMP OF CHICKEN song. That's all. (another band-name pun, like Enigma)
33AnonymousSep 25, 2026 05:15
> While the standard key for that day used rotor order '5-1-2,' MVUEH alone used '2-5-3.'

Doesn't that mean even their own side couldn't read it? lol
34AnonymousSep 25, 2026 05:19
Ah, Komaneci! Takeshi Kitano approves. (Komaneci is comedian Beat Takeshi's signature leg-kick gag)
36AnonymousSep 25, 2026 07:42
Re: #1's AI is already outdated — GPT-6 Sol released on September 22, 2026, so let's have this one crack it again.
37AnonymousSep 25, 2026 08:06
It's Enigma, so it's something you can work out from the hardware gears.
Humans built the thing.
Treating it like some grand mystery is just people killing time.
38AnonymousSep 25, 2026 08:36
Just go ask a German.
40AnonymousSep 25, 2026 08:40
While we're at it, how about some practically useful decoding results too — like freeing up SWF file editing or making it universally compatible.
41AnonymousSep 25, 2026 08:43
News of achievements surpassing human intelligence has really picked up since around July.
We're truly in the middle of the ASI (artificial superintelligence) era.
42AnonymousSep 25, 2026 08:50
Enigma's mechanism has been fully worked out for ages, so the claim that this went unsolved for years sounds fishy to begin with.

Background and Key Points of This Topic

The Enigma cipher is generally considered to have been almost entirely broken during World War II, but in practice, a portion of the huge volume of ciphertext intercepted each day went unread at the time due to limited human resources. This MVUEH message appears to have used rotor order “2-5-3,” differing from that day’s standard order of “5-1-2,” which may be why it couldn’t be solved with the usual key candidates. In the thread, some pointed out that “modern ciphers (AES, elliptic-curve cryptography, etc.) are already at an unbreakable level” — but that’s a comment about modern cryptography, which works on an entirely different principle from Enigma, and some responses conflated the two. It’s also worth noting that the AI’s answer isn’t a simple hallucination: it was judged correct after being cross-checked against already-known ciphertexts and confirmed by an independent third-party researcher, Frode Weierud, who runs the cryptography research site “Crypto Cellar Research.”

*This article is excerpted and summarized from the 5ch (Science News+) thread “[AI] “GPT-6 Astra” Cracks 82-Character Enigma Cipher Unsolved for Years, Even Wrote Its Own Decryption Program.”

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *