| Challenge: | Moderation layers are core component of many products built on user-generated content. |
| Approach: | They propose a system that drafts a content moderation policy based on human-written seed domain information. |
| Outcome: | The proposed system outperforms definition-only and in-context learning baselines on openAI undesired content benchmarks and an in-house multimodal advertisement moderation benchmark. |
Similar Papers
Deep Research with Open-Domain Evaluation and Multi-Stage Guardrails for Safety (2026.acl-long)
Copied to clipboard
Wei-Chieh Huang, Henry Peng Zou, Yaozu Wu, Dongyuan Li, Yankai Chen, Weizhi Zhang, Yangning Li, Angelo Zangari, Jizhou Guo, Chunyu Miao, Liancheng Fang, Langzhou He, Yinghui Li, Renhe Jiang, Philip S. Yu
| Challenge: | Existing deep research frameworks lack adequate evaluation procedures and stage-specific protections. |
| Approach: | They propose a framework with open-domain evaluation and a stage-wise safety benchmark to address this oversight. |
| Outcome: | The proposed framework improves defense success rates by 16.53% while reducing over-refusal rates to approximately 6%. |
Analyzing and Internalizing Complex Policy Documents for LLM Agents (2026.acl-long)
Copied to clipboard
Jiateng Liu, Zhenhailong Wang, Xiaojiang Huang, Yingjie Li, Xiang Li, Chenlei Guo, Xing Fan, Ruhi Sarikaya, Heng Ji
| Challenge: | Large language model agents rely on in-context policy documents to act as effective user assistants. |
| Approach: | They propose an agentic benchmark generator with Controllable Complexity in agent policy across four levels to evaluate agents under increasing complexity. |
| Outcome: | The proposed method outperforms the baseline in data-sparse and high-complexity settings. |
Doc-PP: Document Policy Preservation Benchmark for Large Vision-Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing safety research focuses on implicit social norms or text-only settings, overlooking the complexities of multimodal documents. |
| Approach: | They propose a benchmark to assess the safety of large vision-Language Models (LVLMs) they propose 'Document Policy Preservation Benchmark' to assess document policy compliance. |
| Outcome: | The proposed framework outperforms standard prompting defenses in the evaluation of multimodal documents. |
OmniCompliance-100K: A Multi-Domain, Rule-Grounded, Real-World Safety Compliance Dataset (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing LLM safety datasets rely on ad-hoc taxonomies and lack rule-grounded, real-world cases. |
| Approach: | They construct a rule-grounded, real-world case dataset OmniCompliance-100K from a compliance perspective using a powerful web-searching agent. |
| Outcome: | The proposed dataset spans 74 regulations and policies across a wide range of domains including security and privacy regulations, content safety and user data privacy policies, financial security requirements, medical device risk management standards, educational integrity guidelines, and protections of fundamental human rights. |
VestaBench: An Embodied Benchmark for Safe Long-Horizon Planning Under Multi-Constraint and Adversarial Settings (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Existing safety benchmarks do not represent a diverse range of multi-constraint tasks that require long-horizon planning with a focus on safety. |
| Approach: | They propose a benchmark to assess the safety of embodied AI agents under multiple constraints. |
| Outcome: | The proposed benchmarks show that LLMs perform poorly against their tasks . they also suffer significantly compromised safety outcomes . |
Deep-Reporter: Deep Research for Grounded Multimodal Long-Form Generation (2026.acl-long)
Copied to clipboard
Fangda Ye, Kuicai Dong, Xie Zhifei, Yuxin Hu, Yihang Yin, Shurui Huang, Shikai Dong, Chen Zhang, Jianzhu Bao, Shuicheng Yan
| Challenge: | Recent agentic search frameworks are text-centric, overlooking multimodal evidence . a pressing task is multimodal long-form generation, a new paper argues . |
| Approach: | They propose a unified agentic framework for grounded multimodal long-form generation. |
| Outcome: | The proposed framework is based on a unified agentic framework for grounded multimodal long-form generation. |
ReAlign: Structured Revision for Small Language Model Alignment (2025.findings-emnlp)
Copied to clipboard
| Challenge: | weak policies struggle to generate informative on-policy samples and suffer from unstable gradients when trained on off-police signals from stronger models. |
| Approach: | They propose a training framework that combines stability of on-policy learning with reviser-assisted supervision. |
| Outcome: | The proposed training framework outperforms strong preference optimization baselines on AlpacaEval-2 and Arena-Hard. |
From Off-Policy to On-Policy: Enhancing GUI Agents via Bi-level Expert-to-Policy Assimilation (2026.acl-long)
Copied to clipboard
| Challenge: | Vision-language models are increasingly deployed as computer-use agents that operate desktops and browsers. |
| Approach: | They propose a method that turns static expert traces into policy-aligned guidance . they propose RLVR with a per-task, dynamically updated cache to decompose planning and execution . |
| Outcome: | The proposed model improves UITARS1.5-7B success from 22.87% to 32.13% on OSWorld-Verified and raises a held-out split from 5.74% to 10.30% on MMBench-GUI and Online-Mind2Web. |
Towards Safety Reasoning in LLMs: AI-agentic Deliberation for Policy-embedded CoT Data Creation (2025.findings-acl)
Copied to clipboard
Tharindu Kumarage, Ninareh Mehrabi, Anil Ramakrishna, Xinyan Zhao, Richard Zemel, Kai-Wei Chang, Aram Galstyan, Rahul Gupta, Charith Peris
| Challenge: | Safety reasoning paradigms require high-quality policy-embedded chain-of-thought datasets . generating such data through human annotations is prohibitively expensive . |
| Approach: | They propose AIDSAFE: Agentic Iterative Deliberation for Safety Reasoning . AIDS AFE leverages multi-agent deliberation to iteratively expand reasoning on safety policies . |
| Outcome: | The proposed model improves policy adherence and reasoning quality while maintaining acceptable utility and over-refusal accuracy. |
Inverting the Shield: Systematically Generating Safety Tests from Policy Specifications (2026.acl-long)
Copied to clipboard
| Challenge: | Existing safety evaluation paradigms rely on constructed benchmarks or dynamic red-teaming to probe potential vulnerabilities. |
| Approach: | They propose a framework that integrates specification-based software testing with AI safety. |
| Outcome: | The proposed framework achieves higher coverage and attack success counts compared to baselines. |