ICLR Retracts Oral Acceptance Over Sanctions: Where Is Academic Freedom?

According to a social‑media post, a paper that had been accepted as an oral presentation at ICLR was later desk‑rejected when program chairs noticed the authors listed RAIRI as the institution where the work was performed. Because ICLR is registered in the United States, the organizers said they must obey U.S. law and cannot accept papers from entities on the sanctions list.

RAIRI is the Russian Academy of Sciences’ Institute for Artificial Intelligence, a state‑run research institute based in Moscow that focuses on AI, computer science, and control. In August 2024 the U.S. Office of Foreign Assets Control (OFAC) added RAIRI to the Specially Designated Nationals (SDN) list.

The rejected paper, titled cadrille: Multi‑modal CAD Reconstruction with Reinforcement Learning , proposes a multi‑modal CAD reconstruction model that simultaneously processes point clouds, multi‑view images, and text, ultimately generating executable Python code for parametric CAD models. It uses a two‑stage training pipeline: first supervised fine‑tuning (SFT) on large program‑generation datasets, then reinforcement‑learning (RL) fine‑tuning with online feedback obtained programmatically. The SFT version already outperformed existing single‑modal methods on benchmarks such as DeepCAD, and after RL fine‑tuning it set new state‑of‑the‑art results on ten cross‑modal, cross‑dataset tasks, including real‑world datasets.

The paper entered the review cycle in October‑November 2025, receiving scores of 8, 8, 8, 8, 6 from five reviewers, with a meta‑review explicitly recommending acceptance. On 26 January 2026 the conference announced an Accept (Oral) decision, but on 18 April 2026 the program chairs issued a desk‑reject notice citing the sanctions reason.

Netizens compared this episode with a recent NeurIPS incident where the conference initially barred submissions from Huawei‑affiliated institutions, later apologized and partially rescinded the policy. Some observers note that ICLR’s action appears stricter, while Russian AI researcher @kefirski points out that similar sanctions have affected multiple Russian teams since NeurIPS 2025.

The discussion on social platforms reflects mixed reactions: many express outrage, questioning whether such sanctions harm scientific exchange; others wonder why the authors mentioned RAIRI only on the arXiv preprint and not in the ICLR submission, suggesting possible concealment. In fact, double‑blind conference submissions typically omit institutional affiliations, which are required only in the final camera‑ready version or on preprint servers, where the program chairs identified RAIRI.

Some commenters defend ICLR’s compliance with U.S. law, while others argue that these repeated “academic hegemony” actions erode the principle of open science, prompting a growing sense of resignation among researchers.