Singapore

The proliferation of specialized Large Language Model (LLM) dialogue systems necessitates robust defense mechanisms against unrelated prompts that introduce functional misalignment and unnecessary costs. While LLM-based filtering approaches usually work well, using them requires payments, and that motivated us to develop simple, high-speed, pre-inference input validation techniques. In this paper, we evaluate the efficacy of two semantic pre-filtering strategies applied to a Japanese export control (trade security) application domain: (1) a Centroid-based similarity guardrail (utilizing the mean vector of on-topic embeddings) and (2) a supervised Support Vector Machine (SVM) classifier. Using various multilingual embedding models, we demonstrate that the Centroid-based Similarity Guardrail exhibits superior robustness against adversarial keyword augmentation, effectively maintaining high refusal rates despite injected domain-related terms intended to shift the query’s semantic vector. Furthermore, our analysis of cross-lingual zero-shot transfer confirms that while the strongest multilingual embedding models successfully maintain topic alignment when processing English queries against a Japanese-trained filter, the efficacy is highly model-dependent, underscoring the necessity of model-specific cross-lingual validation for deployment in multilingual environments.

AAAI 2026

Evaluating Lightweight Embedding Guardrails for Cost-Effective Misalignment Mitigation in Export Control Dialog Systems

japanese language

export control

centroid-based classification

topic guardrails

semantic filtering

cross-lingual transfer

adversarial robustness

workshop paper

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.<br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

In recent years, research on role-playing agents that leverage the powerful conversational capabilities of large language models (LLMs) has been actively conducted. Role-playing agents are designed to converse as real or fictional characters, reproducing dialogues that resemble those of the target persona. While many studies have explored how LLMs can generate utterances that reflect a character’s individuality, the examination of utterance generation conditioned on the target character’s emotions has not been sufficiently conducted. Because it is difficult to collect ground-truth data for the target character’s emotions, it has not yet been fully investigated whether conditioning on emotional information can improve the perceived character-likeness of generated utterances. To address this gap, we used scenario data from a visual novel game in which facial expressions, one of the key components of emotional information, are explicitly annotated. We trained an LLM to generate responses conditioned on facial expression labels by incorporating these labels into the input during fine-tuning, and evaluated the resulting utterances in terms of their character-likeness. Our experiments revealed three key findings: (1) conditioning on facial expression labels improves the perceived character-likeness of the generated utterances; (2) providing the LLM with facial expression labels that do not correspond to the character’s actual expressions can still enhance the perceived character-likeness of the generated utterances; and (3) there remains a gap between how humans and LLMs process emotional information in dialogue generation.

Evaluation of Emotion-Conditioned Response Generation for Role-Playing Agents Using Large Language Models: A Case Study with Facial Expression Labels from Visual Novel Data

Data as a Lever: A Neighbouring Datasets Perspective on Predictive Multiplicity

Adapting Multiverse Analysis for Prediction: A Decision-Maker’s Dashboard

On Arbitrary Predictions From Equally Valid Models

Intervention Efficiency and Perturbation Validation Framework: Capacity-Aware and Robust Clinical Model Selection under the Rashomon Effect

Be Intentional About Fairness!: Fairness, Size, and Multiplicity in the Rashomon Set

Evaluating the Ability of Explanations to Disambiguate Models in a Rashomon Set

The rapid proliferation of large language models (LLMs) in applications targeting children and adolescents necessitates a fundamental reassessment of prevailing AI safety frameworks, which are largely tailored to adult users and neglect the distinct developmental vulnerabilities of minors. This paper highlights key deficiencies in existing LLM safety benchmarks, including their inadequate coverage of age-specific cognitive, emotional, and social risks spanning early childhood (ages 0–6), middle childhood (7–12), and adolescence (13–18). To bridge these gaps, we introduce SproutBench, an innovative evaluation suite comprising 1,283 developmentally grounded adversarial prompts designed to probe risks such as emotional dependency, privacy violations, and imitation of hazardous behaviors. Through rigorous empirical evaluation of 47 diverse LLMs, we uncover substantial safety vulnerabilities, corroborated by robust inter-dimensional correlations (e.g., between Safety and Risk Prevention, p = 0.86) and a notable inverse relationship between Interactivity and Age Appropriateness (p = -0.48). These insights yield practical guidelines for advancing child-centric AI design and deployment.

SproutBench: A Benchmark for Safe and Ethical Large Language Models for Youth

Interpretability and robustness remain major challenges for modern Large Language Models (LLMs), especially in settings where conventional evaluation or auditing tools are limited. To address this, we propose Inverse Language Modeling (ILM), a unified training framework that jointly enhances robustness to adversarial perturbations and enables a novel form of gradient-based interpretability. Rather than reconstructing exact input prompts, ILM encourages LLMs to develop gradient-aligned internal representations that allow the model to approximate plausible input patterns underlying a given output. This approximate inversion provides a new mechanism for analyzing model behavior, identifying potential triggers for unsafe generations, and supporting lightweight governance and red-teaming workflows. Our results show that ILM can simultaneously improve robustness and produce meaningful inversion signals, laying a foundation for LLMs that are not only more resilient but also more transparent and analyzable.

Inverse Language Modeling towards Robust and Grounded LLMs

Recent multilingual language models promise support for “100+ languages,” yet speakers of Indigenous and other underrepresented languages still often do not see themselves in these advances. In this work, we take a deliberately simple, secondary-benchmark perspective: rather than proposing a new model or dataset, we re-evaluate an off-the-shelf multilingual natural language inference (NLI) model on public benchmarks that explicitly include Indigenous languages of the Americas. Concretely, we use the AmericasNLI benchmark for ten Indigenous languages and XNLI for English and Spanish, and we evaluate the widely used joeddav/xlm-roberta-large-xnli model under a fixed, zero-shot protocol. Our goal is to answer three questions: (i) How large is the performance gap between high- resource and underrepresented languages under the same model and task? (ii) Are these gaps consistent across languages, or do some communities fare systematically worse than others? (iii) What kinds of qualitative errors arise, and what do they suggest about cultural and linguistic mismatch? Our experiments reveal a striking discrepancy: while English and Spanish reach almost perfect accuracy on XNLI (around 99.8% on our runs), the same model averages only about 43% accuracy across ten Indigenous languages in AmericasNLI, with none exceeding 47%. We also show qualitative NLI failures in Quechua that point to difficulties with morphology, idioms, and discourse-level inference. We argue that even such a simple re-analysis can serve as a low-cost yet high-impact tool for making inequities in multilingual NLP visible, especially for communities that rarely appear in headline benchmarks.

Downloads

Next from AAAI 2026

Evaluation of Emotion-Conditioned Response Generation for Role-Playing Agents Using Large Language Models: A Case Study with Facial Expression Labels from Visual Novel Data

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES