How the famed USENIX Security conf is managing a flood of papers in the AI era

The 35th USENIX Security Symposium (USS) in Baltimore, Maryland has seen an unprecedented number of paper submissions this year, with a total of approximately 3,030 valid papers submitted in Cycle 1 and Cycle 2. This represents a significant increase from the previous year when there were around 2,400 submissions. The symposium chair, Ben Stock, attributes the growth to the broader trend in the security community, citing a similar surge in submissions at the Network and Distributed System Security Symposium (NDSS).

In response to the growing number of submissions, the conference organizers have implemented measures to address potential abuses of AI tools in the research process. According to a transparency report released in January, organizers have developed tools and policies to combat issues such as non-existent references and misuse of AI in the review process.

One such measure involved identifying and rejecting papers with nonexistent references, which impacted 21 out of 1,181 first-round submissions (1.78%). Additionally, over 100 other papers contained at least one reference that reviewers could not confirm.

While the conference has taken steps to tackle AI usage, the primary concern is preserving trust within the security research community. To maintain this trust, organizers emphasize that limited use of AI for polishing human-written text is expected and allowed, even for reviewers up to a certain point. However, usage of AI services to write reviews is strictly prohibited due to concerns over confidentiality violations.

As a result, the conference has asked five out of 496 reviewers to cease participation due to suspected AI use in their reviews.

Despite these efforts, Stock insists that AI-generated submissions have not become a significant challenge for the security community, although he acknowledges that AI may have been used in some submissions.