The Double Life of Code World Models: Provably Unmasking Malicious Behavior Through Execution Traces

Documenti analoghi

• The Good, The Bad, and The Hybrid: A Reward Structure Showdown in Reasoning Models Training
  di: Sahoo, Subramanyam
  Pubblicazione: (2025)
• The Deepfake Detective: Interpreting Neural Forensics Through Sparse Features and Manifolds
  di: Sahoo, Subramanyam, et al.
  Pubblicazione: (2025)
• Calibration Collapse Under Sycophancy Fine-Tuning: How Reward Hacking Breaks Uncertainty Quantification in LLMs
  di: Sahoo, Subramanyam
  Pubblicazione: (2026)
• The Horcrux: Mechanistically Interpretable Task Decomposition for Detecting and Mitigating Reward Hacking in Embodied AI Systems
  di: Sahoo, Subramanyam, et al.
  Pubblicazione: (2025)
• Multi-Agent Systems Execute Arbitrary Malicious Code
  di: Triedman, Harold, et al.
  Pubblicazione: (2025)