Has an AI Just Proved an 18-Year-Old Maths Conjecture?
On 6 November 2026, a company called Physical Superintelligence (PSI) announced something remarkable: its research system, running instances of a non-public Anthropic AI model, had found a proof of the Aaronson–Ambainis conjecture — a problem in mathematics and quantum computing that has been open since 2008. PSI says the proof was complete by the third day of the run, and that the proof-checking program Lean accepted every step. The manuscript and the Lean formalisation are public, but the result has not yet been peer-reviewed or independently verified by outside mathematicians.
What is the conjecture about?
The question is about when quantum computers can be dramatically faster than ordinary computers. Some quantum algorithms, like Shor's 1994 factoring algorithm, are fast because their input has a repeating pattern the algorithm can exploit. The Aaronson–Ambainis conjecture asks about the opposite case: random or unstructured inputs, where no such pattern exists.
Scott Aaronson (University of Texas at Austin) and Andris Ambainis (University of Latvia) conjectured in 2008 that every bounded low-degree polynomial has a highly influential variable — in plain terms, that among all the input bits, at least one of them matters a lot. If true, it would mean that quantum algorithms on random inputs can be simulated by classical computers without an exponential blow-up in effort. That would draw a clear boundary around where quantum speedups are possible.
How was the proof made and checked?
PSI says its system followed a plan sketched by mathematicians Shachar Lovett and Jiapeng Zhang, based on ideas called fractional certificates and level sets, but replaced one step their plan had left as a guess with a brand-new inequality — an "all-pairs inequality" built from a determinant that pairs up input settings one-to-one. The system then improved its own result twice, tightening the final bound, with Lean rechecking every version.
Lean is a proof assistant: a program that checks every logical step of a mathematical argument, the way a spellchecker checks every word. PSI reports that the formal development contains no gaps and no added axioms, and that the proof was replayed through Lean's kernel and through lean4lean, an independent implementation of Lean's type checker.
What still needs checking?
This is where careful reporting matters. Lean guarantees that the theorem as written in Lean is true — but a human reader must still confirm that the Lean statement actually matches the conjecture Aaronson and Ambainis wrote down. The human-readable paper is not itself machine-checked, and the work has not completed peer review. Four Stanford researchers — Guy Blanc, Jordan Docter, Carmen Strassle and Li-Yang Tan, who proved a special case of the problem in August 2026 — have read the proof, sent comments and are preparing a shorter explanation for specialists. In their assessment, published by PSI, they called it "a huge result". But as of the announcement, no assessment by experts outside PSI had been published, and the AI model itself is not public, so nobody can repeat the three-day run.
Why students should care
Whether or not the proof holds up, this is a landmark moment for how mathematics is done. A computer-checked proof can be replayed by anyone, anywhere — the checking is public and mechanical. And the underlying question is a beautiful one for students: it asks, in the end, whether some input always matters more than the rest, and what that tells us about the limits of computers.
Maths rewards the same skills this story shows: breaking a huge problem into steps, checking each step carefully, and being honest about what is proven and what still needs proof.
If your child is working towards GCSE or A-level maths and wants to build that kind of step-by-step confidence, a tutor can help — especially with the algebra, functions and proof techniques that sit behind stories like this one. Our bookshop also has GCSE maths revision workbooks for structured practice at home.
Find the right GCSE maths workbook for your child in our bookshop, or start with a tutor: Book your first lesson
Sources: PSI Reports Machine-Checked Proof of Aaronson-Ambainis Conjecture, Unite.AI; Physical Superintelligence, "A proof of the Aaronson–Ambainis conjecture", 6 November 2026; Scott Aaronson and Andris Ambainis, "The Need for Structure in Quantum Speedups", Theory of Computing 10 (2014)
Curious about maths? Post a request on London Tutors and compare verified maths tutors — online, in person or in a group class.