Paper-Conference

Query Circuits: Explaining How Language Models Answer User Prompts
Beyond Linear Probes: Dynamic Safety Monitoring for Language Models
Towards Understanding Subliminal Learning: When and How Hidden Biases Transfer
Same Answer, Different Representations: Hidden Instability in VLMs
Quantifying the Effect of Test Set Contamination on Generative Evaluations
The Capability Frontier: Benchmarks Miss 82% of Model Performance
Best-of-N Jailbreaking
Context Matters: Analyzing the Generalizability of Linear Probing and Steering Across Diverse Scenarios
Emerging Risks from Embodied AI Require Urgent Policy Action
Establishing Best Practices for Building Rigorous Agentic Benchmarks