USENIX Security conference tackles AI-generated paper submissions with detection measures

The 35th USENIX Security Symposium received a record 3,030 paper submissions, representing significant growth from 2,400 submissions the previous year. While conference organizers attributed part of the increase to AI tools, they report minimal abuse through defensive measures. The conference developed automated tools to detect hallucinated references and rejected papers containing three or more non-existent citations, affecting 1.78 percent of first-round submissions. Over 100 additional papers contained at least one unconfirmed reference. Program co-chairs acknowledged the broader trend of AI-driven submission growth across security conferences, with NDSS submissions jumping from 694 in 2024 to 1,481 in 2026. Organizers also implemented policies against using AI in the peer review process itself, aiming to preserve scientific integrity amid proliferation of readily available large language models.