Cover Image

BARRED: Synthetic Training of Custom Policy Guardrails via Asymmetric Debate

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Mazza, Arnon, Levi, Elad
Format:	Preprint
Published:	2026
Subjects:	Computation and Language Artificial Intelligence Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Lightweight Safety Guardrails via Synthetic Data and RL-guided Adversarial Training
by: Ilin, Aleksei, et al.
Published: (2025)

IntellAgent: A Multi-Agent Framework for Evaluating Conversational AI Systems
by: Levi, Elad, et al.
Published: (2025)

Building a Foundational Guardrail for General Agentic Systems via Synthetic Data
by: Huang, Yue, et al.
Published: (2025)

Intent-based Prompt Calibration: Enhancing prompt optimization with synthetic boundary cases
by: Levi, Elad, et al.
Published: (2024)

MAD-OPD: Breaking the Ceiling in On-Policy Distillation via Multi-Agent Debate
by: Wang, Jianze, et al.
Published: (2026)

SafePred: A Predictive Guardrail for Computer-Using Agents via World Models
by: Chen, Yurun, et al.
Published: (2026)

Why Do Safety Guardrails Degrade Across Languages?
by: Zhang, Max, et al.
Published: (2026)

Noise Injection Systemically Degrades Large Language Model Safety Guardrails
by: Shahani, Prithviraj Singh, et al.
Published: (2025)

Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment
by: Krishna, Kundan, et al.
Published: (2025)

Unified Multi-Task Learning & Model Fusion for Efficient Language Model Guardrailing
by: Neill, James O', et al.
Published: (2025)

Pluralistic Behavior Suite: Stress-Testing Multi-Turn Adherence to Custom Behavioral Policies
by: Varshney, Prasoon, et al.
Published: (2025)

Evaluating the Performance of Large Language Models via Debates
by: Moniri, Behrad, et al.
Published: (2024)

CONSCENDI: A Contrastive and Scenario-Guided Distillation Approach to Guardrail Models for Virtual Assistants
by: Sun, Albert Yu, et al.
Published: (2023)

LoRA-Guard: Parameter-Efficient Guardrail Adaptation for Content Moderation of Large Language Models
by: Elesedy, Hayder, et al.
Published: (2024)

UniGuard: Towards Universal Safety Guardrails for Jailbreak Attacks on Multimodal Large Language Models
by: Oh, Sejoon, et al.
Published: (2024)

Does Training on Synthetic Data Make Models Less Robust?
by: Zhang, Lingze, et al.
Published: (2025)

HAPO: Training Language Models to Reason Concisely via History-Aware Policy Optimization
by: Huang, Chengyu, et al.
Published: (2025)

Provable Length Generalization in Sequence Prediction via Spectral Filtering
by: Marsden, Annie, et al.
Published: (2024)

Identifying Features Associated with Bias Against 93 Stigmatized Groups in Language Models and Guardrail Model Safety Mitigation
by: Gueorguieva, Anna-Maria, et al.
Published: (2025)

Incremental Summarization for Customer Support via Progressive Note-Taking and Agent Feedback
by: Wu, Yisha, et al.
Published: (2025)

Large Language Model Post-Training: A Unified View of Off-Policy and On-Policy Learning
by: Zhao, Shiwan, et al.
Published: (2026)

How Bad is Training on Synthetic Data? A Statistical Analysis of Language Model Collapse
by: Seddik, Mohamed El Amine, et al.
Published: (2024)

XL-Suite: Cross-Lingual Synthetic Training and Evaluation Data for Open-Ended Generation
by: Iyer, Vivek, et al.
Published: (2025)

Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task
by: Li, Kenneth, et al.
Published: (2022)

ThinkGuard: Deliberative Slow Thinking Leads to Cautious Guardrails
by: Wen, Xiaofei, et al.
Published: (2025)

PD$^3$: A Project Duplication Detection Framework via Adapted Multi-Agent Debate
by: Bao, Dezheng, et al.
Published: (2025)

Stabilizing Off-Policy Training for Long-Horizon LLM Agent via Turn-Level Importance Sampling and Clipping-Triggered Normalization
by: Li, Chenliang, et al.
Published: (2025)

Agentic Policy Optimization via Instruction-Policy Co-Evolution
by: Zhou, Han, et al.
Published: (2025)

Towards a Theoretical Understanding of Synthetic Data in LLM Post-Training: A Reverse-Bottleneck Perspective
by: Gan, Zeyu, et al.
Published: (2024)

Can LLMs Speak For Diverse People? Tuning LLMs via Debate to Generate Controllable Controversial Statements
by: Li, Ming, et al.
Published: (2024)

CALICO: Conversational Agent Localization via Synthetic Data Generation
by: Rosenbaum, Andy, et al.
Published: (2024)

A Truly Joint Neural Architecture for Segmentation and Parsing
by: Levi, Danit Yshaayahu, et al.
Published: (2024)

Debate Helps Weak Judges Reward Stronger Models
by: Elasky, Ethan, et al.
Published: (2026)

RigorLLM: Resilient Guardrails for Large Language Models against Undesired Content
by: Yuan, Zhuowen, et al.
Published: (2024)

SIPDO: Closed-Loop Prompt Optimization via Synthetic Data Feedback
by: Yu, Yaoning, et al.
Published: (2025)

EVPO: Explained Variance Policy Optimization for Adaptive Critic Utilization in LLM Post-Training
by: Pan, Chengjun, et al.
Published: (2026)

Predicting Customer Satisfaction by Replicating the Survey Response Distribution
by: Manderscheid, Etienne, et al.
Published: (2024)

Let Models Speak Ciphers: Multiagent Debate through Embeddings
by: Pham, Chau, et al.
Published: (2023)

DiffLM: Controllable Synthetic Data Generation via Diffusion Language Models
by: Zhou, Ying, et al.
Published: (2024)

Virus: Harmful Fine-tuning Attack for Large Language Models Bypassing Guardrail Moderation
by: Huang, Tiansheng, et al.
Published: (2025)