Introduction
As the field of artificial intelligence experiences unprecedented growth, the infrastructure supporting its scientific discourse is under severe strain. Researchers from OpenReview.net and Carnegie Mellon University have recently uncovered a sophisticated "identity theft" scheme targeting the peer-review process of major AI conferences. This investigation reveals how dishonest actors create fraudulent reviewer profiles by impersonating established scientists, ultimately subverting the integrity of the evaluation process to guarantee favorable outcomes for their own submissions.
The Crisis of Scale: Identity Verification
The explosive volume of AI research—with top-tier venues now receiving tens of thousands of submissions—requires the rapid recruitment of massive reviewer pools. Historically, this system has operated on a "trust-based" model. However, as the authors argue, this trust has become a critical vulnerability.
The core challenge lies in identity verification at scale. Program chairs often lack the tools to verify that a person signing up to review is indeed who they claim to be. Because platforms like ORCID do not strictly prevent duplicate profiles or verify personal identities, and many conferences utilize open sign-up forms, the system is susceptible to "sockpuppet" accounts.
Investigative Methodology
This study represents a forensic audit rather than a traditional algorithmic experiment. The research team conducted an in-depth investigation into reviewer profiles across multiple AI conferences managed by OpenReview.net between February 2024 and April 2025. The methodology included:
- Forensic Profile Analysis: Manual investigation of suspicious accounts to identify discrepancies between claimed identities and contact metadata.
- Modus Operandi Mapping: Categorization of the specific techniques used to bypass domain-based verification.
- Systemic Vulnerability Assessment: Analysis of how automated "bidding" and "matching" algorithms were gamed by fraudulent profiles to ensure specific paper assignments.
Mechanisms of Deception
The investigation identified several key tactics used by fraudsters to infiltrate the reviewer pool:
- The Email Alias Trap: Fraudsters exploit university email systems that allow members to create aliases. By creating an address like
[email protected], they can pass domain-based verification while maintaining control over the account to "round-trip verify" a fake OpenReview profile. - Perpetual Invitations: Many conferences automatically import reviewer lists from previous years. Once a fraudulent profile is successfully embedded, it may be perpetually invited to future venues, cementing the fraudster's influence.
- Algorithmic Gaming: Fraudsters do not wait for random assignments; they actively "bid" on their own papers or optimize their fake profiles with specific keywords to ensure matching algorithms pair them with their own submissions.
- Collusion Rings: In some instances, researchers formed rings to trade favorable reviews using these fabricated identities.
Results and Systemic Impact
The audit unearthed 94 reviewer and meta-reviewer profiles confirmed to be fraudulent. Of these, 92 (98%) successfully leveraged reputable university domains (.edu) to establish a veneer of legitimacy. This indicates that institutional email verification is no longer a sufficient proxy for identity.
Proposed Mitigations and Future Work
To combat this growing threat, the authors propose several high-impact reforms:
- Publication Linkage: Requiring reviewers to verify their identity via ORCID or by matching their account email to past publication records.
- Vouching Protocols: Implementing systems where new reviewers must be endorsed by established members of the community, similar to the arXiv model.
- Institutional Reform: Encouraging universities to monitor email alias creation more strictly and cooperate with third-party verification requests.
While OpenReview has begun implementing these safeguards, the authors acknowledge that identity verification remains an ongoing "arms race." Future work will focus on developing "gamification-proof" assignment algorithms and exploring cryptographic verification methods that preserve reviewer anonymity while ensuring accountability.
Summary updated