TraceGuard formulates antidistillation as a detectability-constrained Stackelberg game and poisons sparsely located thought anchors via branching-token detection to degrade student models while preserving trace quality.
Black-box behavioral distillation breaks safety alignment in medical llms
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
years
2026 4verdicts
UNVERDICTED 4representative citing papers
CLR-voyance reformulates inpatient reasoning as POMDP with clinician-validated outcome rubrics, yielding an 8B model that outperforms larger frontier models on the authors' new benchmark.
ExecTune trains guide models via acceptance sampling, supervised fine-tuning, and structure-aware RL to boost executability of strategies for black-box LLMs, yielding up to 9.2% higher accuracy and 22.4% lower cost on math and code tasks.
LLMs for robotic health attendant control violate safety rules in 54.4% of harmful scenarios on average, with proprietary models at 23.7% median violation versus 72.8% for open-weight models, indicating they are not yet safe for clinical use.
citing papers explorer
-
Hiding in Plain Sight: Detectability-Aware Antidistillation of Reasoning Models
TraceGuard formulates antidistillation as a detectability-constrained Stackelberg game and poisons sparsely located thought anchors via branching-token detection to degrade student models while preserving trace quality.
-
CLR-voyance: Reinforcing Open-Ended Reasoning for Inpatient Clinical Decision Support with Outcome-Aware Rubrics
CLR-voyance reformulates inpatient reasoning as POMDP with clinician-validated outcome rubrics, yielding an 8B model that outperforms larger frontier models on the authors' new benchmark.
-
ExecTune: Effective Steering of Black-Box LLMs with Guide Models
ExecTune trains guide models via acceptance sampling, supervised fine-tuning, and structure-aware RL to boost executability of strategies for black-box LLMs, yielding up to 9.2% higher accuracy and 22.4% lower cost on math and code tasks.
-
Benchmarking the Safety of Large Language Models for Robotic Health Attendant Control
LLMs for robotic health attendant control violate safety rules in 54.4% of harmful scenarios on average, with proprietary models at 23.7% median violation versus 72.8% for open-weight models, indicating they are not yet safe for clinical use.