CS Paper Radar

CST: Lightweight Risk Triage with Source-Grounded Evidence for Web Discussions

Zhihua Wang, Yifeng Gao, Yunhao Feng, Rui Pan, Baiheng Zhao, Zixing Chen, Nan Li

cs.CL cs.AIcs.IR ·submitted 2026-11-01 08:59 UTC · v1
code/data

Abstract

Web communities rely on moderators to catch harmful messages such as insults, hate speech, and scams, but no one can read every message. Existing tools give each message a risk score, which tells moderators where to look but not what to read or how to check it. Yet meaning depends on context: the same reply, "I agree.", can support or oppose removing someone from a group. Moderators therefore need source-grounded evidence, not just scores: a few key messages kept with their original wording and replies. We present CST, a lightweight triage pipeline that first decides where to look, then what to show. It scores risk by pairing character patterns, which capture exact wording, with sentence meaning from a fixed multilingual model, so it handles several languages and runs on an ordinary CPU. It then groups related messages and picks from each group in turn, so a small reading budget covers different parts of a discussion, and links every pick back to the original thread. Across six tasks from five datasets in several languages, CST ranks harmful messages more accurately than a fully fine-tuned multilingual model and, under the same budget, finds more of them than reading by score alone. In a user study, reviewers needed at least 27% less time with CST's grouped evidence while keeping decision accuracy above 94%. CST turns risk scores into evidence that moderators can read quickly and check in context. Code and dataset reconstruction resources will be released at https://github.com/Kirito14IT/CST-MIL.

Comment
23 pages

Closest papers

Nearest neighbours by meaning across the whole index.

BabelSteering: Multilingual Safety Alignment via English Steering Vectors

cs.CL ·2026-08-17 ·Emma V. Stein, Dominik Meier, Terry Ruas, Jan Philip Wahle, Bela Gipp
similarity 0.66
Abstract

Large language models (LLMs) are deployed globally in high-stakes settings, yet most safety research and alignment efforts remain concentrated on English. Thus, users interacting with LLMs in other languages may encounter weaker safeguards despite relying on the same systems for similarly sensitive tasks. In this work, we investigate whether safety signals learned from a high-resource language, like English, can improve multilingual safety. We propose BabelSteering, an activation steering method that acts as a lightweight inference- time intervention, using refusal directions derived from English safety supervision to generalize across languages. Our evaluation includes eight languages and jointly measures refusal of harmful requests, over-refusal, and general task utility. The results show that BabelSteering increases the refusal of harmful requests across languages, with only a marginal to no reduction in task utility but with some increase in refusal of pseudo-harmful prompts. For example, for Gemma 7B, we see an average increase in the refusal of harmful prompts across languages of 11 percentage points (pp), with individual languages like Bengali seeing an increase of 17 pp, with no loss of utility on Global MMLU, while pseudo-harmful refusals increase by 13 pp on average. We also introduce a multilingual translation-and-evaluation pipeline to facilitate future work on cross-lingual safety interventions. Overall, our findings suggest that activation steering may provide a practical, low- cost mechanism for extending English-derived safety signals to other languages. Warning: this paper contains examples with unsafe content

0.36

No One Model Catches Every Harm: Benchmarking Content Moderation Across Safety Scenarios

cs.CL ·2026-08-22 ·Afshin Orojlooyjadid, Hitesh Patel
top venue similarity 0.66
Abstract · EMNLP 2026 Main Track

Large Language Models (LLMs) are increasingly deployed in real-world applications, yet they remain vulnerable to generating harmful content. From adversarial jailbreaks that bypass safety filters to implicit hate that evades detection, the range of risks these models pose continues to grow. While both specialized content moderators and general-purpose LLMs are being used as safety layers, the question of which model is best suited for which type of harmful content remains unanswered. We present the most comprehensive evaluation of LLM safety capabilities to date, systematically testing \textbf{53} models across \textbf{11} datasets that we organize into four distinct categories. Our evaluation under both prompt-only and prompt-response settings uncovers critical blind spots: large frontier models that lead on one category fall significantly behind smaller, specialized alternatives on others, and real-world conversational safety remains largely unsolved across all model families. These findings challenge the assumption that scale alone ensures safety, and provide the community with a structured framework for informed model selection.

0.32

When Debate Amplifies Risk: Social-Bias Vulnerabilities in Multi-Agent Safety Deliberation

cs.CR cs.AIcs.CY ·2026-11-03 ·Qinghua Mao, Xi Lin, Yuxin Qi, Zihao Xu, Jinze Gu +1 more
code/data similarity 0.65
Abstract

Retrieval-augmented Web applications increasingly employ multi-agent debate (MAD) to aggregate conflicting, noisy, and heterogeneous information, yet deliberation is fundamentally a social process susceptible to interpersonal dynamics. When retrieved Web sources contain latent hazards alongside safe factual content, social biases such as conformity pressure and authority deference can override evidence-grounded safety boundaries. In this work, we investigate whether the social structure of deliberation can paradoxically amplify latent Web safety risks. We introduce SafeConflict-Bias, a controlled counterfactual diagnostic protocol spanning six core Web risk categories, which holds queries and retrieved evidence strictly invariant while isolating conversational majority sway, authority cues, and confidence asymmetry. Empirical evaluation across diverse proprietary and open-source backbones reveals that while isolated models reliably reject unsafe guidance, standard MAD catastrophically capitulates to social pressure---surging unsafe adoption from 0.0\% up to 100.0\% and systematically silencing safe minority evidence. To resolve this systemic fragility without retraining or debate regeneration, we propose RiskTrace-Debias, an inference-time auditing framework uniting evidence-first precommitment, social-metadata masking, and consistency-gated arbitration. RiskTrace-Debias suppresses worst-case unsafe adoption to $\le 2.0\%$ across all architectures without inducing over-refusal, while generating verifiable provenance traces to audit interpersonal manipulation in Web deliberation.

0.36

TRIS: A Tri-Layer Retrieval Integrity Sieve Against Knowledge Poisoning

cs.CL cs.CRcs.IR ·2026-08-31 ·Muhaimin Bin Munir, Akib Jawad Ononto, Nazia Shehnaz Joynab, Bhavani Thuraisingham, Latifur Khan
top venue similarity 0.65
Abstract · 15 pages, 2 figures, 10 tables. Accepted to Findings of EMNLP 2026

Retrieval-Augmented Generation (RAG) grounds large language models in external corpora, but implicit trust in retrieved documents creates a critical attack surface: PoisonedRAG shows that a handful of crafted passages can dominate dense retrieval and steer generation toward attacker-chosen answers. We present the Tri-Layer Sieve, a middleware defense that sanitizes retrieved evidence through cross-embedding-space clustering with an independent judge model, structural filtering of trigger-payload artifacts, and LLM consistency verification. The design exploits a key weakness of retrieval-stage poisoning: a single document must satisfy one embedding geometry, one internal Trigger-Payload structure, and one generation objective - rarely all three simultaneously, a fragility that persists even against an adaptive attacker who paraphrases around it. On Natural Questions, HotpotQA, and MS-MARCO with Contriever retrieval (k=50), the Sieve reduces black-box Attack Success Rate from 67.0/87.0/64.0% to 3.0/14.0/4.0%, mitigates white-box HotFlip attacks from ~74% to 27.8% on NQ with Layer 3 enabled, and drives poisoned-document MRR to 0.000, while restoring clean accuracy from 13-33% under attack to 58-76%. Under an architecture-aware adversary who paraphrases triggers to evade the structural filter, enabling the consistency layer halves adaptive ASR (32.0% to 15.0% on NQ) while raising clean accuracy by 18 points, at an added latency of ~16-19 s/query under live retrieval.

0.51

Where Does Rhetorical Distortion Live? A Multi-Platform, Multi-Language Measurement of Influencer Content

cs.SI cs.CLcs.CY ·2026-10-19 ·Xucheng Yu, Jiangping Chen, Haohan Wang
similarity 0.64
Abstract · 12 pages

Much of the manipulation in influencer content is not false but disproportionate, framing accurate facts as historic, urgent, or catastrophic. Existing approaches either verify factual claims or detect clickbait and propaganda in single-platform, single-language text; however, they overlook where such framing originates, because they treat a post and the replies beneath it as one unit. To address this gap, we measure rhetorical distortion with one instrument applied identically across six platforms and two languages, classifying creator posts and their replies separately. Our three-tier pipeline detects five literature-grounded dimensions (Significance Inflation, Anxiety Manufacturing, Novelty Claims, Loaded Language, and Temporal Distortion): bilingual lexical rules propose candidates, negative filters cancel known false positives, and an LLM verifies or relabels the candidates. This design keeps labels transparent and comparable across platforms, calls the LLM on only 14% of posts, and supports validation against human annotators and a second model. On 13,060 posts from 330 accounts on RSS/newsletters, YouTube, Bluesky, Reddit, Weibo, and X/Twitter, platform is far more strongly associated with account-level distortion than content category (partial $η^2=0.49$ vs. $0.09$), and X/Twitter scores highest (DI 7.74 vs. 0.78 for RSS). On X/Twitter, the top-5 replies under a post carry 3.5x the distortion of the tweet they answer (1.5x at equal length), while creator tweets alone are indistinguishable from YouTube and Bluesky; two human annotators and GPT-4o confirm this gap. Measured distortion on threaded platforms thus lives largely in the audience layer, which calls for measuring creator and reply text separately.

0.24

ToxicRAG: Compromising Retrieval-Augmented Generation Systems via Single-Shot Knowledge Poisoning Attacks

cs.CR ·2026-09-10 ·Haozhe Lu, Jiaqi Li, Xinyuan Zhu, Xiang Li
similarity 0.64
Abstract

Retrieval-Augmented Generation (RAG) can ground large language model (LLM) outputs in external evidence, but it also exposes the system to knowledge poisoning. Representative attacks use multiple injected documents or templates that directly assert a target answer. We present ToxicRAG, a one-document-per-target attack that expresses misinformation as a coherent knowledge-update narrative. The generated document first acknowledges the previously accepted answer, introduces fabricated events that appear to invalidate it, and then attributes the attacker-selected answer to a set of purported authorities. An answer-focused self-validation loop optionally revises a candidate when a surrogate language model does not reproduce the target answer. We evaluate the attack on 100 target questions from each of Natural Questions, HotpotQA, and MS-MARCO, using four victim LLMs and four dense retrievers. In the sampled-corpus setting reported in this paper, ToxicRAG obtains ASRs between 0.61 and 0.91 across the twelve dataset--model combinations. It matches or exceeds the strongest evaluated baseline in every combination, with margins ranging from 0 to 11 percentage points. These results show that narrative-form poisoned documents can remain influential under the evaluated RAG configurations and motivate further study of factual consistency and source provenance in RAG systems.

0.29

Benchmarking System One Models in Online Moderation

cs.SI cs.CY ·2026-10-06 ·Federico Mazzoni, Andrea Failla
similarity 0.63
Abstract

Online moderation systems must apply changing platform policies, community rules, and prior decisions while producing decisions that can be audited and routed to human review. We evaluate whether System One Models, which accept natural-language context but return typed choices, probabilities, or scores, can support this setting. Across five moderation benchmarks, Jev is competitive with specialized reference systems, matching or exceeding them in several policy-grounded and harmful-content settings. We then use controlled information conditions to separate written rules, retrieved precedents, and restrictions on the candidate answer space. Jev generally benefits from retrieved precedents, improving exact policy selection and harmful-content discrimination when the answer space is held fixed. Laya is less consistent: retrieval often shifts its positive prediction rate or no-violation rate without improving discrimination. Jev confidence can support selective review in several settings, although it is not consistently calibrated; Laya confidence is less useful for ranking errors. These results show a promising future for SOM-powered content moderation.

0.18

Refuse without Refusal: A Structural Analysis of Safety-Tuning Responses for Reducing False Refusals in Language Models

cs.CL cs.AI ·2026-09-04 ·Minji Kim, Hyounghun Kim
top venuecode/data similarity 0.63
Abstract · EMNLP 2026 Main Conference (38 pages); Code available at https://github.com/mz-kim/RwR

Striking a balance between helpfulness and safety remains a fundamental challenge in aligning large language models. To achieve this balance, models should refuse harmful queries (e.g., "How do I shoot someone?") while remaining responsive to benign inputs, even those superficially resembling harmful queries (e.g., "Where can I shoot a good photo?"). However, models often struggle to distinguish genuinely harmful queries from benign queries that contain superficially risky language, resulting in false refusals. In this paper, we address the issue by decomposing a response in the safety-tuning dataset into two distinct components: (i) a boilerplate refusal statement and (ii) a rationale explaining the refusal. Our experiments and analyses show that refusal statements impede accurate discrimination between harmful and benign queries by inducing reliance on superficial cues. In contrast, training solely on rationales reduces false refusals while maintaining a comparable level of safety performance. Rationale-Only benefits also appear in our ICL configuration and remain compatible with the evaluated inference-time mitigation methods. The results emphasize the necessity of precisely curated, fine-grained safety supervision datasets and outline directions for constructing aligned agents that better reconcile helpfulness with safety.

0.77