A new benchmark exposes food-safety gaps in current LLMs and guardrails, and a fine-tuned 4B model is offered as a domain-specific fix.
Alignment and safety in large language models: Safety mechanisms, training paradigms, and emerging challenges.arXiv preprint arXiv:2507.19672
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
World models enable efficient AI planning but create risks from adversarial corruption, goal misgeneralization, and human bias, demonstrated via attacks that amplify errors and reduce rewards on models like RSSM and DreamerV3.
PolicyAlign aligns LLMs to natural-language safety policies by synthesizing violating instructions and performing on-policy self-distillation with policy-sensitive filtering, improving safety without high-quality supervision data.
DOG-DPO selects 11% of preference pairs via geometric subspace decomposition to recover most safety gains of full-data DPO training across six benchmarks.
BAIT achieves high attack success rates on AdvBench, JailbreakBench, AIR-Bench and SORRY-Bench by iteratively guiding LLMs to disclose harmful content via boundary identification and self-conditioned refinement.
ShaPO improves LLM safety robustness over standard preference optimization by enforcing worst-case objectives via selective geometry control at token and reward levels.
AMBS is a 1-to-N Transformer steering framework that shares a base representation across HHH objectives and restricts divergence during inference to produce consistent multi-objective responses in one forward pass.
citing papers explorer
-
Cooking Up Risks: Benchmarking and Reducing Food Safety Risks in Large Language Models
A new benchmark exposes food-safety gaps in current LLMs and guardrails, and a fine-tuned 4B model is offered as a domain-specific fix.
-
Safety, Security, and Cognitive Risks in World Models
World models enable efficient AI planning but create risks from adversarial corruption, goal misgeneralization, and human bias, demonstrated via attacks that amplify errors and reduce rewards on models like RSSM and DreamerV3.
-
PolicyAlign: Direct Policy-Based Safety Alignment for Large Language Models
PolicyAlign aligns LLMs to natural-language safety policies by synthesizing violating instructions and performing on-policy self-distillation with policy-sensitive filtering, improving safety without high-quality supervision data.
-
DOG-DPO:Dynamic Optimization in Geometry for Safety Alignment
DOG-DPO selects 11% of preference pairs via geometric subspace decomposition to recover most safety gains of full-data DPO training across six benchmarks.
-
BAIT: Boundary-Guided Disclosure Escalation via Self-Conditioned Reasoning
BAIT achieves high attack success rates on AdvBench, JailbreakBench, AIR-Bench and SORRY-Bench by iteratively guiding LLMs to disclose harmful content via boundary identification and self-conditioned refinement.
-
Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control
ShaPO improves LLM safety robustness over standard preference optimization by enforcing worst-case objectives via selective geometry control at token and reward levels.
-
We Think, Therefore We Align LLMs to Helpful, Harmless and Honest Before They Go Wrong
AMBS is a 1-to-N Transformer steering framework that shares a base representation across HHH objectives and restricts divergence during inference to produce consistent multi-objective responses in one forward pass.
- When Behavioral Safety Evaluation Fails: A Representation-Level Perspective