A Retrospective on the ICLR 2026 Review Process

2026-03-31 · Source: ICLR Blog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

The International Conference on Learning Representations (ICLR) 2026 received 19,525 valid submissions, with 13,763 ultimately receiving a decision after withdrawals and desk rejections. The conference accepted 5,355 papers, resulting in a 27.4% acceptance rate. This year's program faced significant challenges, including the widespread use of Large Language Models (LLMs) in submissions and reviews, and a security incident involving the OpenReview API. ICLR implemented policies requiring LLM disclosure and used automated detectors to flag LLM-generated reviews and hallucinated references, leading to desk rejections for confirmed violations. The security breach, which exposed author and reviewer identities, necessitated resetting review scores, reassigning area chairs, and extending the meta-review period to preserve review integrity, alongside banning offending individuals.

Key takeaway

For CTOs overseeing research publication or peer review systems, you should prioritize robust security measures and proactive policies for emerging technologies like LLMs. Implement automated detection for content integrity issues, but always back it with human review to manage false positives. Be prepared to execute decisive, transparent remediation plans, such as resetting review stages and reassigning oversight, to safeguard the integrity of your processes during security incidents or widespread policy violations.

Key insights

ICLR 2026 navigated increased submissions, LLM misuse, and a security breach to maintain review integrity.

Principles

• Maintain review integrity under duress.
• Automate detection for scale.
• Human oversight is crucial for flagged content.

Method

ICLR 2026 used dual LLM detectors for reviews, flagged potential issues to area chairs, and employed a system to check references against databases for hallucination, followed by multi-human review.

In practice

• Implement LLM disclosure policies.
• Use automated tools for content detection.
• Reset review processes after security breaches.

Topics

• ICLR 2026 Review Process
• Large Language Models
• LLM Content Detection
• Hallucinated References
• OpenReview Security Incident

Best for: CTO, AI Scientist, Research Scientist, AI Ethicist

Related on AIssential

• CyberThreat-Eval: Can Large Language Models Automate Real-World Threat Research? — Existing LLM benchmarks for CTI lack real-world relevance, necessitating new evaluation methods.
• A review of “Investigating the consequences of accidentally grading CoT during RL” — Accidental CoT exposure during RL training may subtly degrade model monitorability, even if direct impacts are hard to measure.
• Benchmarking Agentic Review Systems — Agentic AI review systems can track human quality judgments and detect errors, showing promise despite improvement needs.
• Research Sabotage in ML Codebases — Detecting research sabotage in ML codebases is challenging for both frontier LLMs and human auditors.
• LLM-as-Judge in Education: A Curriculum-Grounded Marking Pipeline — LLM assessment grounded in official curriculum artifacts improves consistency, transparency, and traceability for high-stakes education.
• ICLR 2026 Response to Security Incident — An OpenReview API exploit led to a major ICLR security breach, compromising anonymity and academic integrity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by ICLR Blog.