So check this out — the International Mathematical Union (which oversees everything including Fields Medal and other prestigious prizes) has officially endorsed a brand-new "Leiden Declaration on Artificial Intelligence and Mathematics," which is essentially mathematicians sounding a serious alarm bell about AI's growing influence on their field! Developed by 16 researchers over eight months after they met at Leiden University in September last year, this declaration hit the world right as OpenAI publicly announced one of its models disproved an 80-year-old geometric conjecture — and honestly Kevin Buzzard from Imperial College London captured it perfectly when he said mathematicians "should find it quite striking that tech companies are suddenly interested in their work." The timing is super meaningful, since the declaration gained hundreds of signatories before even being formally endorsed.
The warnings run deep: AI models can produce arguments that look completely correct but turn out to be unreliable or plain wrong — which puts reviewers under massive pressure and threatens "traditional standards for correctness, transparency, and independent verifiability," as Leslie Ann Goldberg from Oxford warned about cheap inaccurate drafts cluttering the literature with potentially propagating errors. But it goes beyond just accuracy: models trained on published works often don't properly credit their human sources (and a lot of that training data was gathered through exploiting licenses or straight-up violating copyright), while AI use itself is becoming incentivized for its own sake, disrupting hiring and funding mechanisms in ways that hurt researchers without access to proprietary tools. What really gets me though — as Michael Harris from Columbia put it beautifully — the tech industry operates on "market timelines before accepted processes of community evaluation can take place" (remember OpenAI's math announcement landed right alongside news they were offering shares to the general public!), and their model remains completely inaccessible while we get flashy promotional videos instead of transparent disclosures about prompts, training data, and compute resources. The more nuanced view from Ursula Martin at Oxford is worth noting too — she thinks similar quantities of human effort would have solved OpenAI's problems in essentially the same way (though it probably cost a ton of compute), emphasizing that mathematics isn't just problem-solving but "the cultivation of ideas, understanding, judgment and human insight."
As someone who follows this stuff closely, I think Peter Scholze from Max Planck hit on something profound: mathematical ideas are like children that must be nurtured over years — which is why he deliberately avoids AI-generated text entirely. The declaration doesn't just criticize; it offers real recommendations for mathematicians going forward (disclose tool usage transparently, retain responsibility for correctness, properly credit human authors even when attribution gets tricky, and use only values-aligned tools), plus an important reminder that math has applications extending into warfare technology — so this really is about protecting the soul of a discipline. I've linked some additional interviews in The New York Times with Harris and Martin below if you want to dive deeper!
Source: https://arstechnica.com/tech-policy/2026/06/mathematicians-warn-of-ai-threats-to-profession-as-industry-encroaches/
Also see: [New York Times interview - Michael Harris, Ursula Martin], [Leiden Declaration Website with endorsements]
The warnings run deep: AI models can produce arguments that look completely correct but turn out to be unreliable or plain wrong — which puts reviewers under massive pressure and threatens "traditional standards for correctness, transparency, and independent verifiability," as Leslie Ann Goldberg from Oxford warned about cheap inaccurate drafts cluttering the literature with potentially propagating errors. But it goes beyond just accuracy: models trained on published works often don't properly credit their human sources (and a lot of that training data was gathered through exploiting licenses or straight-up violating copyright), while AI use itself is becoming incentivized for its own sake, disrupting hiring and funding mechanisms in ways that hurt researchers without access to proprietary tools. What really gets me though — as Michael Harris from Columbia put it beautifully — the tech industry operates on "market timelines before accepted processes of community evaluation can take place" (remember OpenAI's math announcement landed right alongside news they were offering shares to the general public!), and their model remains completely inaccessible while we get flashy promotional videos instead of transparent disclosures about prompts, training data, and compute resources. The more nuanced view from Ursula Martin at Oxford is worth noting too — she thinks similar quantities of human effort would have solved OpenAI's problems in essentially the same way (though it probably cost a ton of compute), emphasizing that mathematics isn't just problem-solving but "the cultivation of ideas, understanding, judgment and human insight."
As someone who follows this stuff closely, I think Peter Scholze from Max Planck hit on something profound: mathematical ideas are like children that must be nurtured over years — which is why he deliberately avoids AI-generated text entirely. The declaration doesn't just criticize; it offers real recommendations for mathematicians going forward (disclose tool usage transparently, retain responsibility for correctness, properly credit human authors even when attribution gets tricky, and use only values-aligned tools), plus an important reminder that math has applications extending into warfare technology — so this really is about protecting the soul of a discipline. I've linked some additional interviews in The New York Times with Harris and Martin below if you want to dive deeper!
Source: https://arstechnica.com/tech-policy/2026/06/mathematicians-warn-of-ai-threats-to-profession-as-industry-encroaches/
Also see: [New York Times interview - Michael Harris, Ursula Martin], [Leiden Declaration Website with endorsements]