GPT-6 Astra Breaks a Decades-Old Enigma Cipher, AI Builds Its Own Bombe to Crack WWII Code
Artificial intelligence has opened a new chapter in the study of historical cryptography, with advanced AI systems reportedly helping researchers solve two German Enigma messages that had resisted decryption for decades.
The cases involve OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5, but the significance extends beyond the recovery of two historical plaintexts. The experiments demonstrate how modern AI can combine archival investigation, programming, mathematical reasoning, cryptographic analysis and expert feedback within a single research workflow.
One of the messages, known as the MVUEH transmission, was sent by the German Army to the SS-Totenkopf Division on July 10, 1941. It remained unresolved after being made available to cryptographic enthusiasts online in 2005. GPT-6 Astra reportedly located the problem, investigated its historical context, developed its own Enigma simulator and Bombe software, and produced a solution in approximately two days.
A second long-unsolved message was subsequently deciphered with assistance from Claude Opus 5 after cryptanalyst Jack Willis provided the model with additional historical information, including the known signature of a German officer.
The episodes offer an unusually clear demonstration of AI-assisted historical cryptanalysis. They also illustrate an important distinction: these systems did not simply "break Enigma" from scratch. They combined machine reasoning with historical clues, known cryptanalytic techniques, custom software and, in one case, substantial expert guidance.
Why These Enigma Messages Matter
The Enigma machine has become one of the defining symbols of modern cryptography.
Nazi Germany used different configurations of Enigma machines to protect military communications during World War II. The machine transformed plaintext into ciphertext through a sequence of electrical substitutions controlled by components including rotors, a reflector and a plugboard.
Because the machine's configuration changed, successful decryption required more than understanding the basic mechanism. Codebreakers needed information about settings, procedures, operator behavior, message characteristics and, critically, mistakes or predictable patterns that could reduce the enormous search space.
British cryptanalysts ultimately developed sophisticated methods and specialized machinery to attack Enigma traffic. Alan Turing and other researchers played central roles in this effort, while electromechanical Bombes helped automate the search for configurations consistent with intercepted messages.
Yet the historical record contains unresolved fragments.
Some surviving ciphertexts are difficult because important contextual information is missing. Others may contain transcription problems, operational irregularities or insufficient clues. An apparently small historical uncertainty can make an otherwise tractable cryptanalytic problem substantially harder.
That is why solving an individual Enigma message decades after the war can still be technically meaningful.
GPT-6 Astra Took an Autonomous Research Approach
The first recent breakthrough reportedly began when developer Carter Leffen asked GPT-6 Astra to identify an unsolved Enigma message and investigate whether it could be decrypted.
Rather than treating the task as a conventional question-answering exercise, Astra reportedly assembled a broader research workflow.
The model investigated the historical background, identified relevant information, reconstructed the Enigma mechanism computationally and developed software needed to experiment with possible configurations.
This included Python and C++ development for an Enigma simulator and an Enigma Bombe.
That distinction is important.
A language model producing an explanation of how Enigma works is relatively straightforward. Building the software necessary to test cryptographic hypotheses represents a different level of agency.
The AI had to transform historical and cryptographic information into an executable system, use that system to explore possibilities and interpret the resulting evidence.
According to the supplied research account, the complete process took approximately two days.
Veteran Enigma researcher Frode Weierud independently examined the resulting solution and confirmed it. Weierud reportedly characterized Astra's performance as comparable to that of a highly capable professional cryptanalyst and archival researcher, while estimating that equivalent human work could require weeks or months.
The MVUEH Message Had Been Waiting Since 2005
The historical transmission at the center of the first case is known as the MVUEH message.
It was transmitted on July 10, 1941, during the period when Nazi Germany's military communications relied heavily on Enigma encryption. The recipient was the SS-Totenkopf Division.
The message eventually became available to researchers and enthusiasts online in 2005, but publication did not automatically make the underlying cryptographic problem solvable.
This highlights a recurring issue in historical cryptanalysis: information availability and information usability are different things.
Researchers may possess ciphertext without knowing enough about:
The precise machine configuration
Daily key settings
Message procedures
Operator habits
Historical context
Sender and recipient relationships
Potential transcription errors
Relevant German military terminology
An AI system capable of searching across these categories and connecting apparently unrelated clues can approach such problems differently from a human researcher working manually through archival material.
That does not eliminate the need for expertise. Instead, it changes how expertise can be applied.
AI's Real Advantage Is the Combination of Skills
The most interesting characteristic of the Astra case is not necessarily raw computational speed.
It is the integration of several forms of work that traditionally belonged to different specialists.
A conventional cryptanalysis project might require a historian to establish context, a cryptographer to formulate hypotheses, a programmer to implement a simulator, and a researcher to test and interpret results.
An advanced AI agent can potentially move between these roles.
For an Enigma problem, that workflow can look like:
Historical investigation → cryptographic modeling → simulator development → hypothesis generation → computational search → result analysis → independent verification
The ability to maintain continuity across those stages can reduce the coordination burden that traditionally slows research.
This is one of the defining characteristics of agentic AI. The system is not merely generating individual answers. It is pursuing a goal across multiple dependent tasks.
Claude Opus 5 Solved a Second Message With More Human Guidance
The second breakthrough provides an important counterpoint.
Cryptanalyst Jack Willis reportedly used Anthropic's Claude Opus 5 to decipher another previously unresolved Enigma message. Unlike the Astra case, Claude received considerably more assistance.
One important clue was the known signature of a particular German officer.
That information narrowed the historical search space and helped Claude identify the solution.
This difference matters because it demonstrates that AI-assisted cryptanalysis exists on a spectrum.
At one end, an AI system can independently organize much of the research process. At the other, an expert can provide strategically important information that dramatically increases the model's effectiveness.
Neither approach diminishes the underlying achievement.
Instead, it demonstrates that the value of AI may often come from combining machine-scale exploration with human domain knowledge.
Human Verification Remains Essential
The Enigma cases also demonstrate why independent verification remains a critical part of AI-assisted research.
Frode Weierud examined the proposed Astra solution and confirmed it. His decades of interest in Enigma and cryptographic history provided a specialized layer of scrutiny that the model itself could not substitute for.
This is particularly important when AI systems perform historical research.
A model may discover information that appears relevant but has an uncertain provenance. It may also connect clues in ways that are difficult for a researcher to immediately reproduce.
Weierud reportedly raised questions about Astra's references to material associated with a private collection. It was not immediately clear whether the information came from publicly accessible sources, other online material or German archival resources.
That uncertainty illustrates an important principle for AI research: a correct answer is not automatically a fully documented research result.
Researchers need to establish where evidence originated, whether it can be independently reproduced and whether the reasoning connecting evidence to conclusion is sound.
Enigma Is an Ideal Test for AI Research Agents
Historical Enigma problems provide a particularly interesting benchmark for autonomous AI because they combine several disciplines without requiring access to modern classified information.
The machine itself is well documented. Its cryptographic principles are known. The challenge is reconstructing the correct historical circumstances and efficiently searching the remaining possibilities.
That creates a controlled environment for testing AI capabilities.
A capable system must demonstrate:
Historical information retrieval
Technical reasoning
Cryptographic understanding
Software engineering
Search optimization
Pattern recognition
Hypothesis testing
Error correction
Evidence evaluation
This makes unresolved historical ciphertexts useful test cases for agentic systems.
The benchmark is also different from traditional AI evaluations based on mathematics, programming questions or standardized examinations.
A real cryptanalytic investigation has no single predetermined sequence of steps. The researcher must decide what to investigate next.
That is much closer to real scientific and technical work.
The Bombe Was an Early Example of Machine-Assisted Reasoning
There is also a remarkable historical symmetry between today's AI systems and the original Enigma effort.
During World War II, cryptanalysts did not simply rely on human mental calculation. They developed machinery to automate portions of the search process.
The Bombe transformed a repetitive cryptanalytic task into an electromechanical computation problem.
Modern AI systems operate at a radically different technological level, but the underlying idea is familiar: identify the portions of intellectual work that machines can execute repeatedly and allow human researchers to concentrate on higher-level decisions.
The difference today is that AI can potentially participate in those higher-level decisions as well.
A modern research agent can determine which code to write, what experiment to run and which hypothesis to investigate next. That represents a much broader form of automation than the specialized electromechanical systems of the 1940s.
From Codebreaking to Historical Research
The implications extend beyond military cryptography.
Historical archives contain enormous amounts of information that remain difficult to analyze because they are fragmented across languages, documents, databases, handwritten records and specialized collections.
AI systems that can combine archival research with programming could potentially assist with:
Historical cipher analysis
Handwritten document interpretation
Military communications research
Linguistic reconstruction
Archival cross-referencing
Genealogical investigation
Digital humanities
Historical data analysis
The major opportunity is not simply faster searching.
It is the ability to connect different kinds of evidence.
A historical researcher may know that a particular unit operated in a specific region, while a cryptanalyst understands message structure and a programmer knows how to model a cipher. AI systems increasingly have the potential to operate across all three domains.
What These Breakthroughs Do Not Prove
The Enigma cases should not be interpreted as evidence that AI can automatically break modern encryption.
Enigma was designed under historical technological constraints, and its weaknesses have been extensively studied. Modern cryptographic systems are built around different principles and are designed to withstand computational attacks under formally defined security assumptions.
The successful recovery of historical ciphertext therefore does not imply that contemporary encrypted communications are suddenly vulnerable to the same techniques.
Nor does it establish that AI can independently solve every cryptographic problem.
The two reported cases relied on different combinations of automation, historical information and expert input.
The broader lesson is about research capability rather than universal cryptographic power.
The Economics of Autonomous Cryptanalysis
The potential economic impact may ultimately be more significant than the novelty of decoding an old message.
Historical cryptanalysis has traditionally required scarce expert time. A specialist may spend days or weeks reconstructing a problem that ultimately produces a single result.
AI agents could change that equation by allowing researchers to investigate many more candidates simultaneously.
Instead of asking whether a particular problem is important enough to justify weeks of human effort, researchers could increasingly ask whether an AI agent can investigate it at low marginal cost.
That creates a new model of research:
Human selects the problem → AI investigates possibilities → machine performs computational work → human validates the result.
As models improve, the human role may shift increasingly toward selecting meaningful questions, establishing evidence standards and evaluating conclusions.
The Next Frontier: Autonomous Scientific Investigation
The Enigma breakthroughs arrive during a broader transition in AI from conversational assistants to autonomous research systems.
The important capability is no longer simply language generation.
It is sustained goal-directed activity.
An AI agent capable of researching an archive, writing a simulator, running experiments, analyzing results and revising its approach can tackle problems that would be difficult to solve through isolated prompts.
This same architecture could be applied to other domains.
In cybersecurity, AI could investigate historical vulnerabilities and reproduce attacks in controlled environments. In science, it could develop computational experiments. In engineering, it could design and test simulations. In mathematics, it could explore conjectures and generate candidate proofs.
The Enigma examples provide a compact demonstration of this larger transition.
Seven Enigma Messages Still Remain
The work is not finished.
According to the supplied research, seven Enigma messages remain unsolved, alongside another message for which the plaintext is known but the corresponding cryptographic solution remains unresolved.
Those remaining cases provide a natural benchmark for future AI systems.
If increasingly capable agents can solve additional messages while requiring less expert guidance, researchers will gain a clearer picture of how much of the historical cryptanalytic workload can genuinely be automated.
The more difficult question will be whether AI can explain its solutions with enough transparency for historians and cryptographers to reproduce them independently.
Conclusion
The reported decryption of two decades-old Enigma messages represents more than another demonstration of AI's ability to process text.
GPT-6 Astra reportedly investigated an unresolved historical problem, built its own Enigma simulation and Bombe software, conducted computational analysis and produced a solution in approximately two days. Claude Opus 5 subsequently helped solve another message, this time with more substantial guidance from a human cryptanalyst.
Together, the cases demonstrate the emerging power of AI agents that can combine research, coding, reasoning and computation.
They also reinforce the continuing importance of human expertise. Historical context, cryptographic knowledge, evidence provenance and independent verification remain essential for turning an AI-generated solution into a credible research result.
The deeper significance lies in the changing economics of intellectual work. Problems that once required weeks or months of specialized effort can increasingly be assigned to
AI systems capable of working continuously across multiple stages of an investigation.
For Dr. Shahid Masood and the expert team at 1950.ai, the Enigma breakthroughs offer another important indicator of where artificial intelligence is heading. The defining AI systems of the coming era may not simply answer questions. They may investigate them, build the tools required to pursue them, test competing explanations and return with evidence.
The Enigma machine once represented the power of machines to automate cryptanalysis. Eight decades later, AI is beginning to demonstrate something broader, the ability to automate substantial portions of the research process itself.
Key Takeaways
GPT-6 Astra reportedly helped solve the MVUEH Enigma message, which had remained unresolved since it was shared online in 2005.
Astra reportedly developed Python and C++ software for an Enigma simulator and Bombe as part of its investigation.
The reported Astra investigation took approximately two days.
Veteran Enigma researcher Frode Weierud independently examined and confirmed the proposed solution.
Claude Opus 5 reportedly helped cryptanalyst Jack Willis solve another previously unresolved Enigma message.
The Claude case involved more direct expert guidance, including a known officer signature.
Both cases demonstrate AI's ability to combine archival research, cryptanalysis and programming.
Human expertise remained essential for historical context and independent verification.
The breakthroughs do not demonstrate that AI can automatically defeat modern cryptographic systems.
Seven Enigma messages reportedly remain unsolved, along with another whose plaintext is known but whose encryption remains unresolved.
The larger development is the emergence of AI agents capable of pursuing complex research objectives across multiple technical stages.
Further Reading / External References
ChatGPT-6 Astra cracks 85-year-old 1941 Enigma-coded message in two days
PT-6 Astra and Opus 5 crack two Enigma messages that had resisted researchers for decades
Comments