The 35th USENIX Security Symposium, held next week in Baltimore, received a record ~3,030 valid paper submissions this year, up from around 2,400 previously, with organisers acknowledging that wider access to AI writing tools has contributed to the surge. Conference co-chairs say the growth mirrors a broader trend across the security research community, and while AI misuse is evident, defensive measures have kept it from becoming a serious problem so far. This matters because it shows how peer-review systems across academia are adapting to AI-assisted submissions without necessarily descending into crisis, even as scrutiny and workload for organisers increase.

Program co-chairs Ben Stock and Elissa Redmiles detailed in a transparency report how they built tools to catch AI-related abuses, particularly fabricated or "hallucinated" references, by cross-checking citations against sources such as DBLP and arXiv. Papers containing three or more invalid references were rejected outright, a policy that affected 21 of 1,181 first-round submissions (1.78 percent), while over 100 further papers had at least one unconfirmed reference that staff chose not to investigate further to avoid excessive workload. The organisers also flagged concerns about AI being used improperly within the peer-review process itself, though they are wary of over-policing submissions in order to preserve trust within the research community.