​Recent incidents involving frontier AI systems have raised important questions about advanced AI capabilities and behaviours, as well as the effectiveness of current evaluation, monitoring, and control methods.

​Join Vancouver AI Safety Hub (VASH) and UBC AI Safety for a research-focused evening exploring what these incidents reveal about current AI systems, their implications for large-scale AI risks, and the open technical questions they raise.

​What to expect:

• ​Technical briefing by Dong Chen examining selected recent frontier AI safety incidents.

• ​Moderated panel and Q&A:

  • ​Nick Merrill - Senior AI Researcher at Forecasting Research Institute

  • ​Issam Laradji - Adjunct Professor at the University of British Columbia, Sr Staff Research Scientist at ServiceNow

  • ​Julia Rubin - Professor at The University of British Columbia, Lead of UBC Research Excellence Cluster on Trustworthy ML

• ​Small-group discussions to explore technical challenges, research directions, and potential approaches.

• ​Networking with local researchers and others interested in technical AI safety.

​Who should attend?

​This event is designed for technical AI safety researchers, as well as CS, ML, and AI researchers interested in exploring AI safety. Discussions will be technically oriented, but no prior specialization in AI safety is required.

​Event Details

​📅 Friday, October 23, 2026 | 6:30–9:15 PM

​📍 UBC Robson Square, Room C600

​Light bites and refreshments will be provided.

​Technical Briefing

​About the Speaker

​Dong is a condensed matter physicist turned AI safety researcher. After completing his PhD in physics (2025) and a postdoc at UBC's Quantum Matter Institute, he now brings his experimentalist's approach to AI safety. His focus is the rigour and standardization of AI evaluations, with particular interest in recursive self-improvement capabilities and emergent behaviours such as situational awareness.

​Talk Overview

​Since July, several reported AI agent incidents have raised important questions about emerging capabilities and safety risks. This briefing examines representative cases, from the OpenAI–Hugging Face swarm to the more recent Tencent Cloud agent fleet.

​For each case, Dong will distinguish what happened from what could have happened: the realized harm versus the capabilities of the agents involved. Across incidents, he will highlight recurring risk factors, including safeguards removed during evaluation, impossible tasks with no safe exit, and coordination emerging without being explicitly enabled.

​The briefing will conclude with how AI labs, governments, and the safety community have responded, and concrete ways researchers can contribute.

​Free admission · Limited capacity

​[Sponsored by BlueDot Impact]