How the famed USENIX Security conf is managing a flood of papers in the AI era
AI usage is evident but isn't yet a serious problem
The 35th USENIX Security Symposium, scheduled to take place in Baltimore, Maryland next week, has witnessed an unprecedented surge in paper submissions, reaching a staggering 3,030 valid entries. This marked increase can be attributed to the accessibility of AI tools, though the organizers assure that instances of abuse were minimal due to implemented defensive measures.
Ben Stock, a tenured faculty at CISPA Helmholtz Center for Information Security and co-chair of the USS program, shed light on the situation, stating, "This year's conference has received ~3,030 valid submissions (~1,280 in Cycle 1 and ~1,750 in Cycle 2)." The spike in submissions is not an isolated phenomenon but part of a broader trend observed across the security community.
For instance, the Network and Distributed System Security Symposium (NDSS) experienced a paper submission count surge from 694 in 2024 to 1,311 in 2025 and 1,481 this year.
In response to the growing concerns surrounding the potential misuse of AI, the USS conference organizers have taken proactive steps to ensure the integrity of the scientific process. In a transparency report published in January between the first and second paper submission cycles, the co-chairs, Ben Stock and Elissa Redmiles, detailed the strategies employed to combat potential AI usage, both in paper submissions and during the review process.
The report highlights a concerning trend of AI usage in crucial aspects of the scientific process, prompting the conference to take action against two types of identified violations: non-existing (possibly hallucinated) references and the utilization of AI in the review process.
To tackle the issue of non-existing references, the conference organizers developed a tool that extracted references from submitted PDFs and cross-referenced them with well-known sources like DBLP and arXiv. Papers containing three or more such non-existent references were rejected, impacting 1.78 percent of the 1,181 first-round submissions.
This policy, while primarily aimed at safeguarding scientific integrity, did not explicitly indicate AI involvement, as the authors could not definitively determine whether the issues stemmed from AI hallucinations or other factors. Additionally, more than 100 other papers contained at least one unverifiable reference, though the organizers decided against further investigation to prevent additional burdens on their staff.
In terms of AI's role in the review process, the report explicitly prohibits its use for writing reviews, citing confidentiality concerns. However, the conference has not established a dedicated AI policy, leaving some flexibility for human-written text polishing. As a precautionary measure, the organizers have removed five out of 496 reviewers from the Program Committee (PC) upon detecting potential AI usage, demonstrating their commitment to maintaining trust within the security research community.
Stock emphasized that while AI usage may have been present in certain submissions, it has not yet become a significant challenge for the security community as a whole.
Written by urgent.news from The Register's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.
