04
Invited Speakers
Each speaker is anchored to a concrete in-flight or completed experiment on the publication system, not a topical survey.
Academic publishing is becoming AI-native before its governance is ready. This workshop convenes the PC chairs of ICLR, ICML, and CVPR to redesign authorship, peer review, citation, and conference governance under AI.
LLMs and AI agents are no longer just objects of study; they are now participants in the academic pipeline. Authors draft with them, reviewers summarize with them, program chairs deploy them for triage and meta-review. The institutions of science were not designed for this.
From AI for science to AI-native academia. Most AI-for-science venues ask how AI accelerates discovery in biology, chemistry, or medicine. We ask a different question: how should the institutions of science themselves change when AI becomes embedded in authorship, review, citation, publication infrastructure, and scholarly memory?
Failure modes are arising at every node of the pipeline: AI-generated text and fabricated citations at submission, partly or fully AI-generated reviews, hidden prompt injection in manuscripts, score-aggregation mechanisms whose failure modes are not yet understood, recursive feedback into future training corpora. Human-AI co-hallucination (claims no model and no human would produce alone, that arise through their interaction) is one important failure mode among several; this workshop builds a shared taxonomy, security model, and governance framework for the broader transition.
Program chairs and platforms are already running prototype interventions: reviewer-AI-use policies, author self-ranking and Isotonic-Mechanism deployments, in-house AI-review detectors, adversarial-submission audits, full-loop AI-author / AI-reviewer pilots, post-publication citation audits. These are scattered across venues with no shared framework. This workshop is the convergence point.
We invite technical, empirical, and policy submissions on nine topics organized around the AI-native academic pipeline: author → submission → reviewer → AC/SAC/PC → platform → dissemination.
| 1 | AI-Assisted Authorship and Submission Integrityauthor / submitter | LLM use in drafting, related-work synthesis, citation selection, code generation, figure/table generation, polishing, rebuttal, and submission preparation. AI-shaped novelty claims, fabricated or unsupported claims, accessibility for non-native writers, authorship disclosure, submission inflation, AI slop, paper mills, the boundary between legitimate assistance and distorted communication. |
| 2 | Adversarial Manuscripts and Submission-Side Attackssubmission artifact | Hidden prompt injection, invisible text, adversarial figures, malicious metadata, jailbreaks against AI-assisted reviewers, prompt attacks against review assistants, citation checkers, plagiarism tools, and meta-review systems. Document sanitization, submission-portal defenses, review-tool security, red-teaming of AI-assisted conference workflows. |
| 3 | AI-Assisted Peer Review and Reviewer Accountabilityreviewer | Human-only, AI-only, and collaborative reviewing; AI-generated or AI-polished reviews; confidentiality risks from third-party AI tools; reviewer over-reliance; review hallucination; reviewer disclosure policies. Measuring review correctness, helpfulness, specificity, calibration, fairness, and decision impact; tools that support reviewers without replacing human accountability. |
| 4 | Meta-Review, PCs, and Conference-Scale Decision SupportAC / SAC / PC / organizer | AI tools for reviewer assignment, conflict detection, reviewer bidding, desk-reject triage, review-quality monitoring, rebuttal summarization, meta-review synthesis, score aggregation, reviewer-policy enforcement, collusion detection. Mechanism design under AI pressure, including author self-ranking and Isotonic-Mechanism interventions; audit trails for AI-influenced decisions. |
| 5 | Human-AI Co-Hallucination and Cross-Role Error Propagationcross-cutting failure mode | Taxonomies and measurements of false or unsupported scholarly claims that arise through interaction between AI systems and human participants, then propagate through writing, review, rebuttal, meta-review, citation, acceptance, indexing, and reuse. Work that distinguishes co-hallucination from pure model error or pure human error, and identifies intervention points before errors become institutionalized. |
| 6 | Citation, Credit, and Scholarly Knowledge Integritycitation / indexing node | Citation hallucination, citation relevance, support-of-claim verification, citation manipulation, biased citation recommendation, missing-credit detection, bibliometric distortion, citation fairness, knowledge-graph contamination. Methods that verify not only whether a cited work exists but whether it supports the claim being made. |
| 7 | Provenance, Plagiarism, Detection, and Due Processintegrity enforcement layer | AI-generated paper detection extending recent ICLR/ICML audits; plagiarism and paraphrase reuse; watermarking; text, code, and figure provenance; copyright and memorization risks in scholarly writing. Detector reliability, false positives, non-native author harms, and due-process and appeal mechanisms for suspected AI misuse. |
| 8 | Recursive Scholarly Feedback Loops and Corpus Contaminationpost-publication / future corpora | How AI-generated papers, reviews, rebuttals, citations, and summaries enter future scholarly corpora and shape future AI systems. Synthetic-data contamination and model collapse in scientific text, self-reinforcing citation errors, AI-generated review norms, benchmark leakage, homogenization of scientific writing, long-term degradation of scholarly knowledge. |
| 9 | Publication Infrastructure, Policy, and Venue Repositioningplatform / institution / governance | Enforceable AI-use policies for authors, reviewers, ACs, PCs, and platforms; audit trails for AI-assisted decisions; privacy and confidentiality in AI-assisted review; review-data governance; OpenReview, arXiv, ACM, IEEE infrastructure. Scaling review from 15,000 to 30,000+ submissions without collapsing into AI-only evaluation; how top AI venues should reposition authorship, peer review, citation governance, sanctions, transparency, and scientific certification under AI. |
Submission tracks (via OpenReview, non-archival): Short papers (4 pages) and long papers (9 pages), references excluded. Work already accepted at NeurIPS main or other ML venues is excluded.
Each speaker is anchored to a concrete in-flight or completed experiment on the publication system, not a topical survey.
Every panelist has chaired a top AI or medical-imaging venue (NeurIPS, ICLR, ICML, CVPR, MICCAI, MIDL). Humphrey Shi and Mert Sabuncu liaise to the computer-vision and medical-imaging communities.
Moderator: Atlas Wang (UT Austin / XTX Markets), framing the panel around how AI text re-entering the scientific corpus degrades future models.
Panel theme: "Redesigning AI Venues Under AI" (60-min All-PC-Chair Panel + 30-min Open Q&A with the PC chairs).
Panel question: "By NeurIPS 2027, what must AI venues do, operationally, about human-AI co-hallucination across authorship, review, citation, and governance, including AI-review identification, prompt-injection defense, citation grounding, disclosure, and review scaling under 30k+ submissions?"
