Mathematicians are becoming increasingly concerned about artificial intelligence, and with good reason. For centuries, mathematics has been among the most prestigious intellectual disciplines, its practitioners enjoying an authority derived from the supposed certainty of mathematical proof. Philosophers may argue endlessly about truth, historians may revise their interpretations, and scientists may discover that yesterday's accepted theories were mistaken. Mathematics, by contrast, has been regarded as the citadel of demonstrative certainty. Once a theorem is properly proved, it is supposedly settled forever.
Artificial intelligence is now threatening to disturb this comfortable arrangement. The concern is no longer merely that computers will perform calculations faster than human beings. That happened decades ago. The more recent development is that AI systems are beginning to produce original mathematical proofs, tackle problems that have resisted generations of researchers, and generate results at a speed that human mathematicians struggle to assimilate.
The mathematician Terence Tao has warned that mathematics may need to enter a new era, which he calls "Maths 2.0", in which the scarcity of proofs is no longer the defining feature of mathematical research. In an October 2026 article in Nature, Tao argued that AI-generated mathematical results are arriving so rapidly that human understanding is falling behind. There is a genuine danger that mathematical research could become an industrial production line, with machines generating proofs that human beings have neither the time nor the intellectual resources to comprehend.
A group of prominent mathematicians has similarly warned that the objectives of AI companies may be fundamentally misaligned with the deeper purposes of mathematics. Mathematical research, they argue, is not simply about collecting correct answers. It involves conceptual understanding, the development of new methods, and the intellectual growth of researchers. A machine that produces thousands of proofs without contributing to human comprehension might therefore undermine the very activity it appears to advance.
These concerns deserve serious consideration. Yet they overlook an even more disturbing possibility. What happens if artificial intelligence does not merely solve the outstanding problems of mathematics, but begins systematically attacking the foundations upon which mathematics itself has been constructed?
Suppose that a future AI, far exceeding present systems, is directed not towards proving celebrated conjectures but towards discovering inconsistencies, concealed assumptions, conceptual confusions and unjustified transitions in accepted mathematical reasoning. Instead of being asked to solve the Riemann hypothesis or the remaining Millennium Prize problems, it is instructed to examine the logical and philosophical foundations of arithmetic, set theory, infinity, mathematical identity and the accepted principles of mathematical proof.
The results might be considerably more unsettling than another thousand theorems.
There is historical precedent for this kind of intellectual disruption. At the beginning of the twentieth century, mathematicians and philosophers were shaken by the discovery of paradoxes in apparently natural principles of set theory. Bertrand Russell's famous paradox showed that unrestricted assumptions about the formation of sets could lead to contradiction. The response was not to abandon mathematics, but to reconstruct important parts of its foundations through more carefully formulated axiomatic systems.
Kurt Gödel subsequently demonstrated that sufficiently expressive, effectively axiomatized and consistent formal systems cannot establish all arithmetical truths expressible within them. His second incompleteness theorem also imposed severe limitations on such systems' capacity to establish their own consistency. These discoveries did not refute mathematics as a whole, but they demolished ambitious expectations that mathematical certainty could be secured by a straightforward programme of formal proof.
Now imagine an artificial intelligence capable of pursuing this critical tradition with unprecedented power. Rather than accepting established definitions and axioms as unquestionable starting points, it searches for the assumptions upon which mathematical constructions depend. It asks whether a familiar theorem has been interpreted more broadly than its proof warrants, whether an accepted formalisation captures the intended mathematical concept, or whether two supposedly compatible principles generate a contradiction when combined.
Such an AI might discover that a celebrated result depends upon a hidden assumption that mathematicians have failed to notice. It might construct unexpected counterexamples to widely accepted conjectures. It might demonstrate that certain combinations of axioms, previously assumed consistent, are in fact inconsistent. More radically, it might expose fundamental difficulties in philosophical interpretations of mathematics that have become so familiar that they are rarely questioned.
This possibility has particular interest for those who have pursued philosophical criticisms of orthodox mathematics. A small group of philosophers have explored challenges to accepted accounts of mathematical infinity, identity, logical consistency and the philosophical interpretation of mathematical structures. Such criticisms are not themselves established refutations of mathematics, and philosophical objections must be distinguished from formal inconsistency proofs. Nevertheless, they illustrate a research direction that artificial intelligence could eventually pursue on an altogether different scale.
The difference would be one of intellectual power and systematic reach. A human critic might spend years examining a particular mathematical assumption, only to have the criticism dismissed as an eccentric philosophical objection. A sufficiently advanced AI could potentially investigate thousands of related assumptions, construct formal models, search enormous spaces of possible counterexamples, and present its conclusions in machine-checkable form.
Imagine the intellectual consequences if an AI produced a formally verified inconsistency in an axiomatic system regarded as foundational to contemporary mathematics. The discovery would not automatically demonstrate that all mathematics was false. A contradiction in one system might indicate that a particular axiom should be rejected, that a definition was defective, or that the system needed reconstruction. Nor would an inconsistency in a formal theory necessarily undermine every mathematical structure that could be studied independently of it.
But the psychological and institutional consequences could still be enormous. The mathematical community would face a discovery that challenged not merely a particular theorem but the authority of an established framework. Entire research programmes might need reconsideration, and textbooks could require substantial revision. Mathematical results dependent upon the disputed assumptions would need to be carefully separated from those that remained secure.
There is also a deeper philosophical possibility. An AI might not discover a formal contradiction at all, but instead generate powerful arguments that conventional mathematical concepts fail to possess the philosophical significance traditionally attributed to them. It might distinguish formal consistency from truth, mathematical existence from physical existence, or the manipulation of infinite structures from the claim that completed infinities exist independently of mathematical thought.
Such results would be philosophical challenges rather than mathematical refutations in the strict sense. Yet they could be deeply disruptive because mathematicians frequently move between formal constructions and philosophical interpretations without always making the transition explicit.
Consider the familiar assertion that mathematics deals with eternal truths. What exactly does this mean? That a conclusion follows from stipulated axioms? That mathematical objects exist independently of human minds? That mathematical structures accurately describe reality? These are different propositions, and the validity of a formal proof does not automatically settle the philosophical questions surrounding them.
A powerful AI could become an extraordinarily effective instrument for exposing such distinctions. It might produce mathematical results that remain formally impeccable while simultaneously undermining the philosophical interpretations commonly attached to them.
There is an irony here. Much of the current anxiety concerns AI's ability to confirm mathematical expectations by solving problems. But a machine that begins systematically disconfirming accepted assumptions could be far more threatening to professional confidence. It is one thing for a mathematician to discover that a computer has solved a problem before him. It is quite another to discover that a machine has exposed a fundamental error in an intellectual framework to which he has devoted his career.
The response would not necessarily be irrational resistance. Mathematicians would rightly demand rigorous verification, and extraordinary claims about foundational inconsistency would require exceptionally careful scrutiny. AI systems are already capable of producing impressive-looking but defective arguments. A claimed contradiction that depends upon an invalid inference, an inconsistent encoding of the target theory, or an incorrect translation into a proof assistant establishes nothing about the theory being examined.
Nevertheless, suppose these objections were answered. Suppose the AI produced a valid derivation of contradiction from explicitly stated axioms, and independent proof checkers confirmed the derivation. At that point, appeals to professional consensus or institutional prestige would be irrelevant. The mathematical community would have to confront the result.
This suggests that the greatest future contribution of artificial intelligence to mathematics might not be the production of new theorems, but the systematic criticism of mathematical foundations. An advanced AI could function as an intellectual adversary, relentlessly testing assumptions that human researchers have become accustomed to accepting.
Such a development would also transform the relationship between professional mathematicians and philosophical outsiders. Criticisms previously dismissed as speculative could be subjected to rigorous formal investigation. Many would undoubtedly fail, and rightly so. Others might reveal unexpected weaknesses in established reasoning. The authority of an argument would depend less upon the academic status of its author than upon its logical and mathematical merits.
There is no guarantee that this imagined mathematical apocalypse will occur. Contemporary mathematics has already survived profound foundational crises, and its capacity for reconstruction should not be underestimated. Moreover, an AI that exposes the inconsistency of one formal system would not thereby abolish mathematical reasoning. It might instead stimulate the creation of more sophisticated and carefully qualified theories.
Yet the thought experiment reveals a neglected dimension of the AI revolution. The real challenge may not be that machines become better mathematicians than human mathematicians. It may be that machines become better critics of mathematics than its professional guardians.
The mathematicians currently worrying about an avalanche of new proofs may therefore be contemplating the less disturbing half of the problem. A future AI that merely solves their outstanding conjectures threatens their professional role. An AI that systematically exposes defects in accepted mathematical foundations threatens something more fundamental: their confidence in the intellectual architecture they have inherited.
The mathematical apocalypse, if it comes, may not arrive with a machine announcing the solution to the Riemann hypothesis. It may arrive with a machine announcing that some cherished mathematical assumption cannot be maintained, and then supplying a proof that nobody can successfully refute.
That would be an intellectual revolution worth witnessing.