Huy H. Nguyen

Tokyo, Japan

Huy H. Nguyen

AI safety, security, privacy, and multimedia forensics.

Researcher at SB Intuitions · Visiting Associate Professor at the National Institute of Informatics

View CV Google Scholar Publications
Portrait of Huy H. Nguyen

Huy H. Nguyen is a researcher at SB Intuitions, a SoftBank Group company, and a visiting associate professor at the National Institute of Informatics (NII), Japan. His research focuses on AI safety, security, and privacy, with a current emphasis on guardrails and safety evaluation for large language models, vision-language models, and generative AI. Earlier, he worked on deepfake detection, multimedia forensics, adversarial machine learning, and misinformation. He received his Ph.D. from SOKENDAI, in collaboration with NII, in 2022. His work has appeared at EMNLP, AACL-IJCNLP, ICCV, WACV, ICASSP, and IJCB.

Research themes

AI safety & guardrails

Guardrails, safety evaluation, and safety mechanisms for LLMs, VLMs, and text-to-image models.

7 papers

Deepfakes & media forensics

Detecting, localizing, and attributing manipulated and synthetic faces, images, and videos.

28 papers

Adversarial ML & robustness

Attacks and defenses for recognition and generative models, from master faces to backdoors.

28 papers

Misinformation & AI-generated content

Out-of-context detection, fact verification, and identifying or watermarking machine-generated text.

12 papers

Privacy & anonymization

Protecting identity in faces, gait, and location data with learned anonymization.

5 papers

Recent news

  1. SafeGrad accepted to WACV 2027

    On severity-graded safety evaluation, defense, and attack of text-to-image models.

  2. Two papers accepted to AACL-IJCNLP 2026

    On LLM guardrails (The Guardrail Trinity, Main) and black-box backdoor detection (GCoTD, Findings).

  3. Paper on VLM safety neurons accepted to EMNLP 2026

    On neuron-level safety mechanisms across text and image modalities. Main Track.

  4. Panelist at the APO GAIA Public Symposium on AI Readiness

    On safety, talent, and international cooperation, hosted with UNU and the University of Tokyo.

  5. AI Safety: Bridging Academic Research and Industrial Application

    Tutorial at NLP 2026 with Koki Wataoka and Pride Kavumba.

  6. Foundation Models as Guardrails

    Perspective and panel session at APSIPA ASC 2025 with Koki Wataoka. The accompanying paper received the Best Perspective Paper Award.

  7. New publications in disinformation, AI-generated text, and 3D deepfakes

    Research appeared at ICMI, IJCNN, IJCB, the WACV Workshop, and in leading journals.

  8. Three papers on deepfake detection accepted to IJCB 2024

  9. BTAS/IJCB 5-Year Highest Impact Award

    For the 2019 BTAS paper on multi-task deepfake detection and segmentation.

  10. Two papers on generative and adversarial robustness accepted to ICIP 2024

    On spurious feature generation and defense against physical adversarial attacks in the infrared domain.

Selected publications

  1. WACV 2027

    SafeGrad: A Severity-Graded Benchmark for Safety Evaluation, Defense, and Attack of Text-to-Image Models

    M.-D. Vo, J. Li, H. H. Nguyen, P. Kavumba, and K. Wataoka

    Winter Conference on Applications of Computer Vision (WACV) 2027

  2. EMNLP 2026

    Do VLMs Share Safety Neurons Across Modalities?

    J. Li, J. Zhang, M.-D. Vo, H. H. Nguyen, P. Kavumba, and K. Wataoka

    Empirical Methods in Natural Language Processing (EMNLP) 2026, Main Track

  3. AACL-IJCNLP 2026

    The Guardrail Trinity: Procedural Safety Schemas, Reasoning Chain Decomposition, and Dual-Mode Inference for LLM Guardrails

    H. H. Nguyen, P. Kavumba, J. Li, M.-D. Vo, and K. Wataoka

    Asia-Pacific Chapter of the Association for Computational Linguistics and International Joint Conference on Natural Language Processing (AACL-IJCNLP) 2026, Main

  4. APSIPA ASC 2025 Best Perspective Paper Award

    Foundation Models as Guardrails: LLM- and VLM-Based Approaches to Safety and Alignment

    H. H. Nguyen, P. Kavumba, T. Kurosawa, and K. Wataoka

    Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) 2025, pp. 2564–2569

  5. WACV 2023

    Closer Look at the Transferability of Adversarial Examples: How They Fool Different Models Differently

    F. Waseda, S. Nishikawa, T.-N. Le, H. H. Nguyen, and I. Echizen

    Winter Conference on Applications of Computer Vision (WACV) 2023, pp. 1360–1368

  6. IEEE T-BIOM 2022 Journal

    Master Face Attacks on Face Recognition Systems

    H. H. Nguyen, S. Marcel, J. Yamagishi, and I. Echizen

    IEEE Transactions on Biometrics, Behavior, and Identity Science, pp. 398–411 (2022)

  7. ICCV 2021

    OpenForensics: Large-Scale Challenging Dataset for Multi-Face Forgery Detection and Segmentation In-The-Wild

    T.-N. Le, H. H. Nguyen, J. Yamagishi, and I. Echizen

    International Conference on Computer Vision (ICCV) 2021, pp. 10117–10127

  8. AINA 2020

    Generating Sentiment-Preserving Fake Online Reviews Using Neural Language Models and Their Human- and Machine-based Detection

    D. I. Adelani, H. Mai, F. Fang, H. H. Nguyen, J. Yamagishi, and I. Echizen

    International Conference on Advanced Information Networking and Applications (AINA) 2020, pp. 1341–1354

  9. ICASSP 2019

    Capsule-Forensics: Using Capsule Networks to Detect Forged Images and Videos

    H. H. Nguyen, J. Yamagishi, and I. Echizen

    International Conference on Acoustics, Speech, and Signal Processing (ICASSP) 2019, pp. 2307–2311

  10. BTAS 2019 BTAS/IJCB 5-Year Highest Impact Award

    Multi-task Learning for Detecting and Segmenting Manipulated Facial Images and Videos

    H. H. Nguyen, F. Fang, J. Yamagishi, and I. Echizen

    International Conference on Biometrics: Theory, Applications and Systems (BTAS) 2019, pp. 1–8

All 72 publications

Academic activities

Service

Reviewer

Conferences: NeurIPS, ICLR, ICML, CVPR, ECCV, WACV, ICME, ACL RR, APSIPA ASC.

Journals: IEEE Access, TIP, TIFS, IEEE/CAA JAS, ACM TOMM, Elsevier, EURASIP JIVP, and IEICE.

Chairing

Session chair

  • APSIPA ASC 2020: Deep Generative Models for Media Clones and Its Detection.
  • APSIPA ASC 2023: Multimedia Security and Privacy in the Age of Deep Learning.

Community

Organizer & instructor

  • LAVA Workshop at ACCV 2024 and ACM MM 2025.
  • APSIPA ASC 2026 Winter School 2: Foundation Models and Agentic AI.

Get in touch

Research conversations welcome.

For research collaborations, talks, or academic inquiries, please contact me at: