No. 031

Normative and infrastructural responses to AI in mathematics are developing at different speeds. The normative questions (credit, disclosure, governance of tool adoption) have no community consensus yet, even as individual researchers and small collectives are publishing position pieces faster than institutions can respond. Infrastructure is accumulating in parallel: a community GitHub repository to revise OpenAI's mathematical proofs, a proposed attribution framework for AI-consumed scholarly content, algorithmic auditing that found more than 10,000 conflict-of-interest violations in peer review by scanning half a million papers. The Simons Foundation's $15M initiative, co-funded by Anthropic and OpenAI, tries to address both, funding community workshops on AI evaluation standards while also providing researcher access to the AI tools those standards are meant to govern.

Mathematics and AI

• Simons Foundation Launches $15M Effort To Help Mathematicians Navigate AI Era

  Simons Foundation, October 15 2026

  The foundation administers an open call for proposals, with Anthropic and OpenAI contributing tools and compute alongside cash, creating a governance arrangement where the two companies providing the tools are also co-funding the community standards meant to govern them.

• There's never been a better time to go into math

  3Blue1Brown, October 16 2026

  A 40-minute talk by Grant Sanderson works through AI as a proof machine, conjecture machine, and theory machine, then addresses what follows if AI becomes capable of most mathematical work, making this the most concrete publicly available framing of AI's effects on mathematical research careers aimed at those entering the field.

• oai-math-revised: a community hub for revising OpenAI's math write-ups

  Terence Tao's blog, October 24 2026

  A community GitHub repository invites mathematicians by specialty to annotate and revise OpenAI's October mathematical write-ups, addressing incomplete citations and unclear proofs, providing crowd-sourced quality infrastructure for AI-generated results that have no formal peer review process.

• Recommendations on the use of AI for PhD students and supervisors

  Proofs and Prompts, October 19 2026

  A collective of symplectic geometry researchers published practical guidance for doctoral researchers and supervisors, with the central principle that researchers must retain authority over which questions matter and must verify all AI-generated results themselves rather than treating model output as reliable.

Authorship and Accountability

• On disclosure and shame

  Proofs and Prompts, October 24 2026

  Max Lieblich (University of Washington) argues that framing AI use as something to disclose encodes shame into the norm itself, and proposes treating AI contributions like human ones: coauthorship when the contribution is substantive, acknowledgment when it is assistive.

• Thousands of research studies violate journal policies on collaborations between editors and authors

  Retraction Watch, October 22 2026

  EditorCheck scanned 512,000 papers from PLOS and MDPI and found more than 10,000 cases where the handling editor had recently coauthored with one of the paper's authors, demonstrating that algorithmic auditing can make systematic conflict-of-interest screening practical at scales no manual process could reach.

Tools and Evaluation

• New paper: Emerging trends in AI agent evaluation

  Normal Technology, October 19 2026

  A paper from Kapoor, Narayanan, and colleagues identifies 33 trends in how AI agents are evaluated and argues that useful evaluations must examine full agent decision trajectories and sociotechnical context, not only final outcomes, which is also the gap between current agent benchmarks and actual research deployment.