I wrote a reflection on @COLM_conf 2025 — two takeaways that stuck with me months later:
1. Safety isn't just about blocking "how to build a bomb." Carlini made a compelling case that subtle, long-term harms matter more.
2. New interaction paradigms (generative interfaces) = new attack surfaces. Presented by @Diyi_Yang
Full Post: https://t.co/hK8BQxiPgx
📅 See the full list of upcoming talks & seminar details:
https://t.co/vZNlrbYE9R
Stay tuned for announcements on individual sessions and other AI Security Lab news!
#AI#Security#Privacy#Seminar#GenerativeAI#UMass
🎙️ AI Security Seminar is back!
After a successful first semester featuring speakers from academia and industry, we’re excited to launch a new semester of engaging talks on cutting-edge topics in AI security.
📝 Register to attend & receive updates:
https://t.co/OORfe7hqLM
Registering ensures you receive the Zoom link for each talk and our lab’s updates. Videos of all talks will not necessarily be published, so registration is the best way to stay connected.