When AI Goes Live: Security and Privacy Failures We Didn’t See Coming

As AI systems become part of higher education operations and research, security and privacy risks often appear after deployment. This session explores real-world AI failure patterns such as insecure defaults, over trusted model outputs, data drift, and misuse of AI pipelines. Drawing on hands-on experience building AI systems in regulated environments, the session connects technical issues with institutional responsibility. It shows why security by default and privacy by design must be built into systems from the start. Attendees will gain practical insight into identifying risks early, aligning governance with implementation, and building AI systems that protect trust while supporting innovation.