Xiao Zhan

Emerging risks from agentic AI demand urgent interdisciplinary collaboration to secure our future.

UBD reveals that traditional accuracy metrics can mask critical sample-level discrepancies in model behavior, leading to a more nuanced understanding of data contamination effects.

Unlock robot safety reasoning: a new dialogue interface lets humans ask "why," "why not," and "what if" about robot safety interventions, grounded directly in the robot's internal constraint-based decision-making.

Romantic AI platforms treat your deepest secrets as reusable training data, granting themselves broad rights while leaving you holding the bag.