I still remember the moment the first reports hit my screen. Hundreds of mathematical manuscripts, all generated by an AI model no one outside the company can touch, dropped into a public repository in a single day. Some of the claims touched problems that have frustrated human researchers for generations. Within hours mathematicians were openly stunned. Within a day crypto researchers started talking about something they called bunker mode. That sequence of events feels less like incremental progress and more like a sudden shift in the ground under our feet.
When AI Starts Solving Problems Humans Have Stared At for Decades
On a quiet early October day the company released 722 manuscripts covering 372 groups of related findings. The model had been tested on roughly four thousand research problems. Each result in the final collection used, on average, computing power equivalent to about three hours of advanced thinking mode. The papers touched prime numbers, geometry, and the mathematics that sits underneath computing itself. Some claims reached into territory that has resisted some of the world’s best minds for a very long time.
Perhaps the most eye-catching claim involves the Riemann hypothesis. That puzzle dates back to 1859 and concerns the hidden patterns that govern prime numbers, the building blocks from which every whole number greater than one can be constructed. Mathematicians study those patterns through a special function and look especially carefully at the points where it equals zero. The full hypothesis predicts that all the important zeros fall along one precise line. The AI did not claim to have proved that entire statement. Instead it claimed to have ruled out zeros across an enormous, permanently defined region where earlier proofs had offered a much narrower guarantee.
For specialists the new dividing line sits at a value of 7/8 while the full hypothesis requires 1/2. The related equations are known as the Riemann zeta function and Dirichlet L-functions. For everyone else the practical point is simpler. The model claims to have established a far stronger limit on where these crucial mathematical points can appear, without claiming to have settled the entire puzzle. If the claim holds up under independent scrutiny it would count as a major advance toward understanding one of mathematics’ most famous unsolved problems.
Other papers claim progress on the Hodge conjecture, which concerns the structure of complex geometric shapes, though only for a class of objects called CM abelian varieties. The language is dense, the statements careful, and the potential implications large. Yet the sheer volume of the release created an immediate problem. How do human researchers even begin to check hundreds of advanced results that arrive all at once?
Mathematicians React With Awe and Alarm
One prominent mathematician reacted with pure astonishment. He noted that a human who produced a comparable quasi-Riemann result would be an instant candidate for the highest honors in the field. Previous bounds on the location of zeros had grown thinner the higher one moved up the imaginary axis. The new claim offered something stronger: a genuine zero-free strip. That single detail alone would have been considered a massive breakthrough if a human research group had announced it after years of careful work.
If a human did this, it would be an instant Fields Medal, no questions asked.
The reaction was not uniformly celebratory. Some researchers pointed out that the model remains a black box controlled by one company. Others noted that the release followed an earlier announcement claiming progress on the Navier-Stokes problem, the mathematical laws governing moving fluids. That earlier claim had already sparked disputes over verification and credit. The pattern was becoming clear. Results of potentially historic importance were arriving faster than the community could comfortably digest them.
I’ve found that the emotional temperature in these discussions often reveals more than the technical details alone. Awe is real. So is the sense that the usual pace of discovery has been disrupted. When human understanding is no longer the primary bottleneck, questions of process, access and responsibility rise quickly to the surface.
Calls for a Boycott and Competing Views on Progress
The Association for Human Mathematics urged researchers to stop collaborating with the company. Their argument was straightforward. Releasing more than seven hundred files at once was not a demonstration of scholarship but a demonstration of power. The group rejected the claim that the release advanced the field in a healthy way and called for a return to research centered on human understanding.
A separate advisory group had issued guidelines only days earlier. Those guidelines asked AI laboratories to stop testing advanced mathematical problems on proprietary models that the research community cannot access. They also outlined a process for responsibly publishing results already generated, including proper citations, formal verification where feasible, independent repositories and support for human researchers trying to understand the work. After the release the same group emphasized that consultation did not equal endorsement. The publication was treated as the beginning of scientific scrutiny, not the end.
Another public commentator offered a sharply different interpretation. He accused parts of the scientific community of standing in the way of progress and suggested that precarious institutional arrangements had long held back faster advances. The tension is unmistakable. On one side stands the desire to protect the integrity of mathematical practice and the centrality of human insight. On the other stands the view that delaying the release of powerful new tools is itself a form of obstruction.
In my experience these debates rarely resolve cleanly. They tend to evolve as the results themselves are checked, revised or withdrawn. That process is already under way.
Corrections, Withdrawals and Formal Verification
The release has already seen adjustments. A sign error in one paper’s proof invalidated that paper along with two others built on its construction. All three, involving Weil classes and K3 surfaces, were withdrawn. Fourteen other manuscripts were revised to repair proofs, correct statements or clarify assumptions. References were updated in thirteen more. The repository now lists 719 manuscripts.
According to the company, the main results in roughly three hundred of the remaining papers have been translated into Lean, specialized software that checks whether each step of a mathematical proof follows logically from the assumptions. The software verifies the statement it is given. Mathematicians still have to decide whether that statement accurately captures what the paper claims to prove. For results without such computer checks the company itself acknowledges that some may contain mistakes.
This mixture of rapid generation, partial formal verification and ongoing human review creates an unusual situation. Durable advances may sit side by side with claims that will need further revision or outright withdrawal. Sorting one from the other will take time, expertise and coordinated effort across the community.
Why Crypto Researchers Started Talking About Bunker Mode
The implications did not stay inside university mathematics departments. An Ethereum Foundation researcher publicly urged the blockchain industry to begin calmly planning for what he called bunker mode. His core concern is that advanced AI could eventually discover a mathematical shortcut for breaking the digital signatures that protect major cryptocurrencies, potentially before the powerful quantum computers that security researchers have long worried about arrive.
Every conventional cryptocurrency wallet relies on a secret private key to authorize transactions and a related public key to verify them. Finding the public key from the private one is easy. Reversing the process is designed to be practically impossible with current technology. The fear is that an advanced AI might uncover a classical shortcut that makes that reversal feasible on ordinary computing hardware. In a worst-case scenario the widely used ECDSA signature system could become vulnerable in months rather than years. The researcher offered that scenario as a possibility to prepare for, not as a demonstrated breach.
Today I call upon the blockchain industry to calmly begin planning for bunker mode.
His practical recommendation centers on a controlled mass migration of assets to fresh addresses whose public keys remain hidden behind a hash. Holders, especially large and sophisticated ones, should consider moving the bulk of their funds to addresses that have never signed a transaction. When they do sign one, remaining funds should also move to a new address, possibly generated from the same seed phrase. The advice comes with repeated warnings against rushing. A botched migration can cause real losses more quickly than the hypothetical attack it is meant to prevent.
For some Bitcoin address types an unused address conceals its public key behind a cryptographic hash until its funds are spent. That feature provides a degree of protection that disappears once the address is used. The recommendation is therefore not a universal fix for every wallet or address format, but a targeted hardening step that requires no new cryptography and no new wallet software.
Mixed Reactions Across the Crypto Community
Reactions varied. One co-founder of a major blockchain project agreed that rapid advances in AI-powered mathematics deserve serious attention, yet cautioned that poorly executed wallet migrations can produce immediate and irreversible losses. A head of cryptography at a large exchange stated there was no evidence whatsoever that the mathematical foundations of elliptic-curve security had been weakened. The original researcher went further still, noting the striking under-representation of cryptographic breakthroughs among the released results and suggesting that government intervention in related academic work is not unheard of. He offered no direct evidence of intervention in this particular release.
Perhaps the most interesting aspect is the difference in time horizons people are willing to entertain. Some treat the risk as still distant and theoretical. Others argue that recent mathematical surprises, including unexpected disproofs of long-held conjectures, should make the industry more open to the possibility of classical counterparts to known quantum algorithms. Elliptic curves carry rich algebraic structure. That richness has enabled sophisticated techniques over the years. The same structure might also leave room for unexpected classical attacks once mathematical superintelligence becomes a practical tool.
Hashes, by contrast, are deliberately designed to minimize algebraic structure. That difference leads some researchers to favor hash-based cryptography as a longer-term defensive posture. The idea is to rely on a single battle-tested hash family and avoid structured mathematical assumptions from curves, lattices or isogenies wherever possible.
Practical Steps Suggested for Asset Holders
The concrete advice circulating in these discussions focuses on prevention rather than panic. Large holders whose public keys have already been exposed are encouraged to lead by example. Trackers of exposed public keys already exist for certain assets. Institutions holding significant cold storage are being urged to review their exposure and consider gradual rotation to addresses that keep public keys hidden until necessary.
- Move bulk funds to never-used addresses where public keys remain behind a hash
- When a transaction is required, also move remaining funds to a fresh address
- Avoid rushed migrations that can introduce human error and permanent loss
- Consider rotating keys more frequently for high-value signers such as oracles or security councils
- Explore multi-signature arrangements that include hash-based schemes where appropriate
Smaller holders are sometimes described as enjoying partial cover simply because large numbers of historically exposed addresses already exist. That cover is imperfect and should not be mistaken for strong security. The broader message is that preparation can begin without new cryptographic primitives and without waiting for a confirmed break.
Exiting bunker mode safely will eventually require post-AI cryptography. Some researchers argue for accelerating work on hash-based approaches and end-to-end formal verification. Existing roadmaps that already lean in that direction may need to be revisited in light of the speed of recent mathematical progress.
The Deeper Questions the Release Leaves Unanswered
Two questions hang over the entire episode. Can independent mathematicians verify discoveries that arrive faster than they can reasonably review them? And if increasingly powerful AI begins changing fields as consequential as cryptography, who decides how that knowledge is tested, shared and put to use?
The model that produced the work remains proprietary. Selected reasoning summaries have been released and support for workshops and further verification has been promised. Still, the broader mathematical community faces the practical task of distinguishing lasting advances from claims that will need revision. That task is made harder by the volume of material and by the fact that the underlying system is not available for independent experimentation.
In the crypto world the conversation has already moved beyond pure theory. Whether or not a classical break of elliptic-curve signatures is imminent, the psychological and operational shift toward greater caution is real. Asset holders are being asked to treat mathematical superintelligence as a factor that can compress timelines previously measured in decades into much shorter windows.
I keep coming back to the human element in all of this. Mathematics has always progressed through a mixture of sudden insight and painstaking verification. When the sudden insight arrives in bulk from a system no outsider can interrogate, the verification step becomes both more urgent and more difficult. Crypto security has always rested on the assumption that certain mathematical problems remain hard for the foreseeable future. If that assumption starts to look less permanent, the practical response is not panic but deliberate hardening of the systems that hold value.
Looking Ahead Without Losing Sight of Verification
The coming months will test whether the most ambitious claims survive independent scrutiny. Some results will almost certainly stand. Others will be refined or withdrawn. The formal verification already applied to a substantial fraction of the papers offers one useful filter, yet human judgment about the meaning of those formal statements remains indispensable.
For the blockchain industry the prudent path appears to be calm preparation. Controlled migration of high-value assets to addresses that keep public keys hidden, more frequent key rotation for critical signers, and accelerated research into hash-based alternatives all represent steps that can be taken without assuming the worst-case scenario has already arrived. The goal is resilience rather than overreaction.
What makes this moment distinctive is the compression of timescales. Mathematical progress that once unfolded over generations is being produced, at least in draft form, in a matter of days. The institutions that evaluate, publish and apply that progress were not designed for that speed. Adapting them while preserving standards of rigor is the real challenge facing both pure mathematics and the applied fields that depend on it.
Whether bunker mode becomes a temporary posture or a longer-term operating assumption will depend on what independent researchers ultimately confirm. Until those confirmations arrive, the combination of awe at the mathematical claims and caution about their downstream effects seems like the only rational stance available. The ground has shifted. The work of understanding exactly how far, and in which directions, has only just begun.
In the end the story is not only about machines producing proofs. It is about the people who must decide what those proofs mean, how they should be shared, and what practical defenses make sense while the dust settles. That human responsibility has not disappeared. If anything, it has become more visible and more urgent than before.
