Unprecedented Paper Submissions
The 35th USENIX Security Symposium (USS) commencing next week in Baltimore is witnessing an unprecedented volume of paper submissions. Although AI tools have contributed to this increase, the conference organizers believe that instances of misuse were limited due to effective safeguards. However, they are also maintaining discretion to preserve trust within the security research community.
Ben Stock, a seasoned faculty member at the CISPA Helmholtz Center for Information Security and co-chair of the USS program, revealed to GadgetLad that the conference received around 3,030 valid submissions (~1,280 in Cycle 1 and ~1,750 in Cycle 2). In comparison, last year saw a mere 2,400 submissions. Stock indicated that this pattern is apparent throughout the security community, with NDSS experiencing an increase in submissions from 694 in 2024 to 1,311 in 2025 and 1,481 this year. “Not exactly unprecedented,” Stock commented, yet they have expanded their Program Committee (PC) to accommodate the influx.
Image: An Influx of Papers
Identifying Questionable AI Practices
An April paper titled “More Versus Better: Artificial Intelligence, Incentives, and the Emerging Crisis in Peer Review,” points out a 42% rise in submission volumes at major journals since ChatGPT emerged in 2022. In the USENIX Security ’26 transparency report, Stock and co-chair Elissa Redmiles, a computer science expert at Georgetown University, outlined their strategy for detecting AI-related misconduct in submissions and reviews.
The report emphasizes the community’s recognition of LLMs (Language Model Machines) in writing and coding tasks. However, there are concerns regarding AI infiltrating the scientific process. They focused on two primary problematic behaviors: fabricated references and AI involvement in reviews. After identifying and rejecting a paper with false references, they developed a tool to verify references against major databases such as DBLP and arXiv. Papers containing three or more fake references were disqualified — impacting 21 out of 1,181 first-round submissions (1.78%).
Image: AI Illusions
Establishing Boundaries for AI Use
Over 100 additional papers included references that reviewers could not verify. Some may represent false positives due to typographical errors, so they refrained from extensive investigations to avoid overwhelming the staff. A firm boundary has been set against using AI for bibliography preparation, fearing potential threats to scientific integrity. Nevertheless, a light application of AI for refining human-written text is permissible, even for reviewers, as long as it remains regulated.
“We haven’t implemented a strict AI policy,” Stock remarked, “but we have informed PC members that AI-generated reviews are unacceptable, primarily due to confidentiality violations.” They have identified several individuals utilizing AI and subsequently removed those PC members, allowing affected authors to resubmit. Five of 496 reviewers were affected by this policy.
Stock believes that AI-generated submissions are not yet a significant concern for the security field. However, the extent of AI involvement behind the scenes remains uncertain.
Image: The AI Balancing Challenge
Summary: Paper Triumph or AI Triviality?
It seems AI is asserting its influence in academia, but USENIX is actively pushing back. With their defenses in place, they are addressing AI’s cunning tactics while remaining open to some AI assistance when appropriate. Let’s observe how this intriguing technological drama unfolds next week!
