Claude Fable 5.1 solved a 370-year-old cipher in forty-four minutes.
This tracks with what I wrote about AIs doing mathematics: It’s good at things that involve lots of searching and testing.
EDITED TO ADD (9/14): I’m not sure if this result is correct. The blog comments have more discussion.
KC • September 9, 2026 9:56 AM
To add to @foo’s comment, it’s interesting that Wikipedia (currently without a source) seems a little confabulated on this finding. I’m also not sure what author it’s referring to.
“After inspecting the original 1653 book in the British Library, the author revealed that the two-line numerical cipher does not appear after the 32nd petition. Instead, the text simply concludes with decorative flourishes, Latin phrases, and a closing statement.”
However, the Vals AI blog seems to reference a source copy at Edinburgh: Maitland Club, 1834.
So, am wondering if more historical cryptanalysts will weigh in?
Clive Robinson • September 9, 2026 10:45 AM
@ Bruce, ALL,
With regards,
“It’s good at things that involve lots of searching and testing.”
As far as “Current AI LLM Systems” are concerned all the “big achievements” we have been informed about are achieved with intensive “searching and testing” gone through rapidly has been the foundation.
If we look at these achievements they are somewhat niche and based around a very large problem domain that can be checked with simple rules. With the testing effectively independent and thus easily parallelizable.
This usefulness even though niche was in effect withheld from the public by the AI companies, untill someone got upset and over a weekend churned out “OpenClaw”, and agenetic use of AI rather rapidly hit the big time.
The thing to note is that “searching and testing” has to be easily parallelizable in some way. If not agenetic use of LLMs will not be particularly fast.
It’s something you would have thought would be obvious to anyone, yet even though it is, and has been pointed out, all the AI Hype hooplah has blinded most to it…
Which makes me wonder how much else that is obvious about AI is being missed due to the AI company hype.
One thing to consider is Current AI Agenetic ability whilst useful, and in some cases is very advantageous, is very niche… Which is not something “crazy mad” investors “throwing money into a hole in the water” are going to want to hear as it makes the ROI tiny.
Clive Robinson • September 9, 2026 11:24 AM
@ Bruce, ALL,
With regards your article in The Guardian where you say,
“We think the contrary view is more likely, at least in the short-term. AI models are nowhere near as capable as experienced academic mathematicians.”
I have to disagree.
It will not be “at least in the short-term” it will be effectively for ever. Because that “gap” between “Current AI LLM Systems” and “experienced academic mathematicians” is not going to close appreciatively.
The reason as with any other “force multiplier tool” we’ve created is that it will “free up drudge work” and alow humans to have ability and time to upscale their creativity.
Thus tools like “LEAN”[1] which has taken a lot of the drudge out of “proofs” will take a lot drudge out of hypothesis testing. Enabling a human mind more time to do the creative work of seeing the loose threads that lead onto new and very different hypothesis, that fall well outside the capabilities of the stochastic fuzzing found in LLMs.
I’ve talked about this issue with “Current AI LLM Systems” in the past, in that they can work their way close in to “known Classes” and find new Instances there. But they can not due to the way they work actually find new instants that form new classes that are more than a short distance away from a known instance. Thus the do not take “intuitive but directed leaps” as humans can just “drunkards walks” at best (though they can as tools test those human inspired intuitive leaps).
[1] LEAN is Open Source and under fairly rapid development. It is based on the “Calculus of Inductive Constructions”(Coq)
https://en.wikipedia.org/wiki/Lean_(proof_assistant)
https://en.wikipedia.org/wiki/Calculus_of_constructions
As I’ve mentioned before another area to keep your eye on is other types of logic that can be more amenable to tool use, such as “Computability Logic”(CoL)
https://en.wikipedia.org/wiki/Computability_logic
They are all methods that can be automated into tools thus act as “force multipliers”.
Rontea • September 9, 2026 11:53 AM
What stands out here is the disciplined process: identify a tractable target, leverage the internal structure of the document, and then verify against historical context. Fable 5.1 didn’t just throw compute at the problem—it recognized a pattern humans overlooked.
For cybersecurity and intelligence folks, this is a reminder that many longstanding puzzles are solvable not by brute force, but by patient, structured reasoning. Once automation can focus human attention on the right clue, entire classes of “unsolved” challenges may collapse quickly.
lurker • September 9, 2026 2:15 PM
@too, KC
Perhaps Claude’s hallucination rate should be part of the record?
The name Fable for the machine may be an appropriate slip, a Wikipedia editor laments the use of chosun.com as the source for the decoding claim. I tried a ddg search for sources, the only recognisable one that turned up was msn.
The Urquhart original was printed before the Statute of Anne 1710, before the Licensing of the Press Act 1662, so ripoff copies may well exist with dodgy additions.
And my local radio news this am leads with a scary story from Anthropic: 10% chance AI will eliminate humanity in next decade
lurker • September 9, 2026 2:21 PM
@Clive Robinson, ALL
Related news: Open AI claims ~10,000 agents took 88 hours to solve the Navier-Stokes equations. Again there is backlash and a claim of cheating.
Ferentarius • September 9, 2026 2:24 PM
A 10% chance of annihilation is but the statistical whisper of the despair that has animated humanity since its first thought. We birth machines in the hope they will exalt us, yet secretly we crave their betrayal—because only in the shadow of our erasure do we feel the weight of existing. The apocalypse was never a question of if, but of when our ingenuity would finally exhaust our desire to persist.
kjfriedo • September 10, 2026 2:57 AM
A useful update to the source question raised above: Fable’s Distich mechanism is reproducible against the 1834 Proquiritation ordering, and a surviving 1653 copy at the National Library of Scotland (H.32.a.39) is documented with the Distich present. The BL/EEBO witness lacks it, so the discrepancy is copy-specific, not a disproof of the cipher.
We have also now obtained research scans from Glasgow’s physical 1652 The Jewel (Sp Coll Bi2-l.17), confirming that the Cyphral Octastich leaves are integral to that copy. The remaining task is coordinate-by-coordinate verification of the proposed Octastich solution.
The interesting lesson is that AI may have found a real structure, but historical cryptanalysis still needs copy-specific provenance and reproducibility.
kjfriedo • September 10, 2026 3:09 AM
One further source-level update: Glasgow ASC has now confirmed that its 1652 copy of The Jewel (Sp Coll Bi2-l.17) is complete at the end and that the leaves containing the Cyphral Octastich are integral to the volume.
The supplied scans also show continuous pagination through pp. 127–130 and 149–160, so the proposed one-page shift from position 159 cannot simply be explained by a missing or duplicated page in this witness.
The remaining question is now strictly cryptanalytic: does the rule reproduce the claimed plaintext coordinate by coordinate from the physical 1652 text?
r • September 10, 2026 3:24 AM
@lurker, feren
bbc last night brought up the point that the 10% claim is from a midlevel developer and that other “higher ups” (certain ceo’s) put that risk at 20-25%.
“cross” training of biological data makes me particularly nervous, we might incentivise hemorrhagic fever or aerosolization/droplet properties with one model and then take the force feedback resulting instance and use it to engineer a cancer treatment.
we REALLY need people who understand quarantining software to be doing this work but they’re prioritizing “fast and loose” to keep the gas running.
kjfriedo • September 10, 2026 3:49 AM
One further update: I have now checked the critical part of Fable 5.1’s Cyphral Octastich solution against research-quality scans of the physical 1652 Glasgow copy, Sp Coll Bi2-l.17.
The strongest result is that positions 149–157 reproduce exactly the same unresolved sequence reported by Vals/Fable:
CONERTHTO
Position 158 still works with the normal page rule, and from position 159 onward the proposed one-page shift is strongly supported by the next letters:
159 -> H
160 -> I
161 -> S
giving the beginning of THIS USURP’D AUTHORITIE.
There is one caveat: coordinate 159 depends on how a line-broken word on p.158 is counted, so the shift is strongly supported but not perfectly clean.
The physical scans also show continuous pagination through the critical ranges, so the shift cannot simply be explained by a missing or duplicated page in this copy.
This moves the Octastich result beyond a merely plausible AI-generated plaintext: the mechanism is independently supported by the physical 1652 witness in the most difficult test region. A complete 285-coordinate physical replication is still pending.
anonymous • September 16, 2026 12:32 PM
It seems to me that “AND CONERTH TO THIS USURP’D AUTHORITIE
REIGNE IN HIS ROYAL PREDECESSORS STEAD” makes sense. Conerth being an alternate spelling of Cornerth/Cornert, an offshoot of the de Grey family prominent at that time.
Subscribe to comments on this entry
Sidebar photo of Bruce Schneier by Joe MacInnis.
foo • September 9, 2026 9:08 AM
Not sure what’s going on here:
https://github.com/reticuli-labs/panel-artifacts/blob/main/distich-refutation-2026-09-01/FINDINGS.md