Computer science conferences are adopting submission and reviewer limits to manage heavy review workloads, but the available conference reports do not show that generative AI alone caused the increase in submissions. The rules vary: some cap papers listing an author, while others limit reciprocal reviewing. Their effects on review quality and fairness have not yet been established.
Why are computer science conferences limiting submissions?
Organizers say they are trying to keep peer review workable as submission volumes grow and reviewer capacity comes under pressure. They also cite concerns about research quality, integrity, and fairness. AI-generated or low-quality submissions are part of some organizers’ stated concerns, but the figures available do not isolate AI as the cause of rising submission volume.
As an Amazon Associate I earn from qualifying purchases.
The scale is clear in AAAI’s report on its 2026 Main Technical Track. On August 25, 2025, AAAI said it had received almost 29,000 submissions. After removing papers that violated submission policies—including the author submission cap—roughly 23,000 remained under review, nearly twice the number of papers reviewed by AAAI-25. AAAI also reported that almost 20,000 submissions came from China. These are AAAI’s counts for that cycle, not an independently audited or sector-wide measure.
How do the current caps differ?
The policies below apply to particular conferences, editions, or tracks; they are not a common rule across computer science. A cap on papers naming an author is also different from a limit on how many submissions rely on someone to review.
#1 Best Overall
| Venue and cycle | What is limited | Ceiling and enforcement | Stated AI policy |
|---|---|---|---|
| ICLR 2027 | Submissions listing an author | Up to 20 papers for an author; one paper for a “new author,” defined by the Program Chairs as someone who has never published at a major AI, machine-learning, computer-vision, robotics, or NLP conference. The policy announcement does not specify a surplus-paper enforcement mechanism. | The Program Chairs cite low-quality and AI-generated content among their concerns; the announcement does not establish a general permission or ban for AI assistance. |
| OSDI ’26 | Submissions listing an author | Eight per author. The official call does not state an enforcement mechanism in the cited policy. | Work wholly or largely generated by AI is treated as fraud; AI editing tools are acceptable. This is specific to OSDI ’26. |
| SIGCOMM ’26 | Submissions listing an author | If an author appears on more than eight submissions, surplus submissions will be desk-rejected. | The call directs authors to ACM’s generative-AI policy, including disclosure of AI-created content. |
| ICML 2026 | Reciprocal-review assignments | A cap applies to the number of papers that can designate the same person as a reciprocal reviewer; the call does not state a numeric ceiling here. | Generative AI tools, including LLMs, may assist with writing or research. |
| NeurIPS 2026 Position Paper Track | AI use in submitted papers | Not stated in the track’s AI-use statement. | Papers must be substantially human-written; AI use is restricted to copy-editing or similar peripheral changes. This rule is for the Position Paper Track, not all NeurIPS submissions. |
ICLR’s Program Chairs said the 20-paper ceiling could reasonably have been set lower: “We personally find it difficult to imagine how one author could contribute meaningfully to 20 papers, and think this number could have defensibly been chosen to be much smaller.” Their post also says about 20% of ICLR 2026 submissions had no reciprocal reviewer, based on comparing author lists with the final program-committee list. That is the organizers’ described calculation, not a general estimate for other conferences.
Did AI cause the surge in conference papers?
The evidence cited by these organizers does not establish that. AAAI reported a very large submission load, and ICLR’s Program Chairs named AI-generated content among the concerns behind their policy. Those observations support the conclusion that AI is part of the discussion about review pressure, not that it accounts for a measured share of submission growth. No comparable multi-conference dataset in these reports attributes the increase to generative AI.
It is also important to distinguish AI rules from capacity rules. ICML 2026 permits generative AI assistance while limiting reciprocal-review assignments. OSDI ’26 permits editing tools but calls work wholly or largely generated by AI fraud. NeurIPS 2026’s restriction applies to its Position Paper Track. These approaches cannot be collapsed into a single policy for “computer science conferences.”
Do submission caps make peer review fairer?
That remains an open question. A cap may reduce the number of papers attached to prolific authors or encourage more reciprocal reviewing, but enforcement can create difficult choices about which submissions to reject. A 2025 arXiv paper proposes a fairness-aware framework for desk rejection under submission limits; it is an analytical proposal, not evidence that caps have reduced reviewer hours or improved fairness in practice.
ICLR’s policy illustrates the trade-off organizers are trying to manage. Alongside the 20-paper author limit, it gives a stricter one-paper ceiling to authors meeting its “new author” definition. The Program Chairs discuss reviewer capacity and fairness concerns, including the risk of disproportionately burdening junior researchers in large groups. Whether the rule distributes opportunities more fairly will depend on its implementation and effects, which the policy announcement alone cannot demonstrate.
Quick Recap
Best Value
What authors should check before submitting
- Read the call for the exact edition and track. OSDI ’26’s rules, for example, should not be assumed to apply to every USENIX event, and NeurIPS’s cited AI restriction is track-specific.
- Check how the venue defines an author or reviewer. ICLR 2027’s new-author definition depends on prior publication at specified kinds of major conferences; ICML’s rule concerns reciprocal-review assignments rather than an author-listed-paper ceiling.
- Confirm the consequence of exceeding a limit. SIGCOMM ’26 explicitly says surplus submissions will be desk-rejected. Do not assume another venue uses the same enforcement.
- Follow the venue’s AI-use and disclosure rules separately. A submission cap does not tell authors whether AI assistance is permitted, restricted, or subject to disclosure.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
