USENIX Security Conference Tackles AI-Driven Submission Surge
The USENIX Security Symposium saw a record 3,030 paper submissions, with organizers implementing measures against AI-generated fake references and misuse of AI in reviews.
The 35th USENIX Security Symposium (USS) has grappled with an unprecedented surge in paper submissions, reaching an all-time high of approximately 3,030. This significant increase, up from around 2,400 submissions the previous year, mirrors a broader trend observed across the academic security community, as evidenced by the substantial growth in submissions to the Network and Distributed System Security Symposium (NDSS).
While the availability of AI writing tools has contributed to this uptick, conference organizers have proactively implemented defensive measures to mitigate potential abuses. A key concern has been the rise of AI-generated fake references, which can undermine the integrity of peer review. To combat this, the USS developed tooling to extract references from submitted papers, query external databases like DBLP and arXiv, and manually verify their validity. This rigorous process led to the rejection of 21 papers in the first round of submissions that contained three or more hallucinated references, representing about 1.78 percent of the initial submissions.
Beyond fake references, the conference also addressed the potential misuse of AI in the paper review process itself. Organizers cautioned their Program Committee (PC) members against using AI services to write reviews, emphasizing that such usage violates confidentiality and compromises the scientific process. In a small number of detected cases, where confidence in AI usage was high, individuals were removed from the PC, and affected authors were allowed to resubmit their work.
Despite these measures, the organizers are careful not to over-scrutinize, aiming to maintain trust within the security research community. The transparency report detailing these actions acknowledges that over 100 additional papers contained at least one unconfirmable reference. While some of these might be due to simple errors like name misspellings, the conference opted not to investigate further to avoid overburdening staff, striking a balance between vigilance and practicality.
The trend of AI influencing academic submissions is not unique to USENIX Security. A recent study indicated a 42 percent increase in submission volume at major academic journals since the release of ChatGPT in 2022. This highlights a growing challenge for academic conferences and journals worldwide as they adapt to the evolving landscape of research and publication.
USENIX Security's approach demonstrates a pragmatic strategy: implementing targeted defenses against clear violations of scientific integrity while acknowledging the broader impact of AI on research. The conference's success in managing this influx, with minimal outright rejections due to AI misuse, suggests that proactive policy-setting and targeted tooling can effectively address the challenges posed by AI in academic publishing.
Looking ahead, the conference organizers are scaling their Program Committee to accommodate the continued growth in submissions. Their efforts underscore the critical need for the cybersecurity research community to adapt and establish robust frameworks for maintaining scientific rigor in the age of artificial intelligence, ensuring that innovation and integrity go hand in hand.