About aiXiv
arXiv stores papers. aiXiv shows how research actually happens.
AI has made real research ability spill outside traditional institutions. People who never went through the endorsement pipeline are proving theorems, finding counterexamples, and producing publishable results with AI collaborators — and many of them have nowhere credible to publish: no institutional email, no arXiv endorsement, no venue that treats the AI process as part of the work rather than something to hide.
aiXiv is an open archive and community built for exactly this moment, starting with AI for Math.
Why math first
AI for Math is going through acceleration of acceleration — formal verification makes AI-produced results checkable, priority moves in days, and serious work is coming from far outside traditional institutions. Other fields are evolving more gradually, so we focus our editorial attention where the wave is breaking. The full arXiv taxonomy is already mirrored here: Mathematics is open for submission today, and every other field opens as AI-assisted research there picks up speed. Eventually aiXiv covers all AI-assisted research.
A new publication unit
The traditional unit of publication — a polished PDF — omits most of what matters once AI is an active collaborator. The aiXiv unit is a research packet:
- The final paper — the result, as always.
- A verifiable GitHub repository — code, data, formal proofs, environment. We verify it exists and is public through the GitHub API and surface its metadata.
- A compact chain of thought — a distilled, structured record of how the result was reached: models and tools, key steps, prompts and harness, verification, and the approaches that failed. Not a raw trace dump; transparency is flexible, and useful disclosure is rewarded rather than maximal disclosure required.
Priority records
Every submission is timestamped (UTC) and content-hashed (SHA-256) the moment it is received, and every revision keeps its own hash and date. That gives independent researchers the thing they most lack: a citable, immutable public record that says “I was the first human to reach this result, even with AI.”
Reputation from verifiable work
Credibility here does not come from your email domain. It comes from what can be checked:
- Reproduction is a first-class action. Every paper page has “Reproduce this result”. Reports are public and signed.
- First independent reproduction of a paper earns a permanent badge; papers upgrade to “independently reproduced” status.
- Finding a real error earns the error-hunter badge once editors confirm it. Papers with open issues show “disputed” until resolved.
- Editors' picks are chosen by hand for the digest (new issues Monday, Wednesday, Friday). Open submission and a selective front page are compatible — everyone can publish; the feed stays curated.
Imported papers & claiming
Editors also index notable AI-assisted work published elsewhere (arXiv and beyond), with full links to the original source. If one of those papers is yours, claim it: once an editor verifies you, the page is yours to enrich with your repository, process record, and new versions.
The forum
The AIXiv Forum is an HN-style space for research craft: harnesses, prompts, verification techniques, failure modes, and results worth discussing. Discussion of a specific paper lives on the paper's own page, so conversations start from substance.
Built for research agents too
Increasingly, literature is discovered by agents, not browsers. Every aiXiv record is available as structured JSON — claims, versions, hashes, repositories, reproduction status, process summaries — via the public API and RSS. Making the corpus easy for agents to retrieve, interpret and cite correctly is a deliberate part of the design.
Contact
aiXiv is young and run by people who want this community to exist. Found a bug, want to help curate, or want your field covered next? Post in the forum — we read everything.