Security Science Lab
The Security Science Lab at Fujitsu Research India conducts research at the intersection of Artificial Intelligence, Security, and Trustworthiness, with the goal of building AI systems that are secure, resilient, transparent, and trustworthy in real-world environments.
As AI becomes increasingly integrated into enterprise systems, information ecosystems, autonomous agents, and physical environments, new security and trust challenges are emerging. AI systems can themselves be attacked or manipulated, while AI-generated and AI-amplified content can be used to deceive people, spread misinformation, and undermine trust in digital information.
Our lab develops fundamental AI science and practical technologies to understand these emerging threats, build robust defences, and enable trustworthy deployment of next-generation AI.
Mission
To build AI systems that are secure, resilient, transparent, and trustworthy in real-world environments by advancing research at the intersection of AI, security, and trustworthiness.
Research Philosophy: From Attack to Resilience
A central philosophy of the lab is that robust defences require understanding how AI systems fail under realistic adversarial conditions.
Our research follows an integrated methodology:
Discover → Attack → Understand → Defend → Stress-Test → Deploy
We develop adaptive attacks to expose vulnerabilities, analyse the mechanisms underlying model failures, design principled defences, and stress-test those defences against stronger and evolving adversaries. Our goal is to move beyond systems that merely appear secure toward AI that is genuinely resilient.
Research Areas
- AI & LLM Security
We study vulnerabilities in Large Language Models and generative AI systems, including backdoors, adversarial prompting, adaptive attacks, secure model adaptation, and robust defence mechanisms.
- Secure RAG & AI Agents
We investigate security threats arising from knowledge-intensive AI, including adversarial retrieval, knowledge injection, latent and semantic attacks, secure RAG, and trustworthy AI agents.
- AI Safety & Alignment
We develop methods for RL-based safety alignment, reward modeling, concept-aware alignment, preference optimization, and robust safety mechanisms, with an emphasis on simultaneously preserving safety and utility.
- Misinformation & Information Integrity
Our research focuses on AI methods for detecting, explaining, and mitigating misinformation across languages, modalities, and information ecosystems, including multilingual, multimodal, and explainable misinformation detection.
- Deepfakes & Synthetic Media
We investigate image, video, audio, and multimodal deepfake detection, synthetic media forensics, cross-modal consistency, and robust detection under evolving generative models.
- Watermarking & Content Provenance
Our research advances technologies for AI-generated content watermarking, content attribution, provenance, authentication, and tamper detection, enabling trustworthy digital content ecosystems.
- Multimodal & Physical AI Security
We study security and robustness of multimodal foundation models, vision-language systems, embodied AI, autonomous agents, and physical AI, including attacks that propagate from the physical environment through perception and reasoning to action.
- Explainable & Trustworthy AI
We create approaches for explainability, interpretability, concept-based reasoning, AI auditing, behavioural analysis, and trustworthy decision-making, enabling AI systems whose behaviour can be understood and evaluated.
Research Themes
Our broader research vision is organized around four interconnected themes:
- AI Safety and Security
LLM security, Adversarial ML, Backdoors, RAG security, AI agents, Physical AI- RL-based alignment
- Information Integrity
Misinformation, Multilingual AI, Multimodal misinformation, Deepfakes, Synthetic media, Provenance
- Trustworthy AI
Explainability, Interpretability, AI auditing, Watermarking, Robustness, Responsible deployment
Key Achievements & Recognition
- Security Science Lab received FRIPL Appreciation Award 2026 for its contribution in AI trust and social trust technologies.
- Received the Global Fujitsu Award for work on technologies addressing disinformation and AI risks.
- The Security Science Lab has filed six patents covering key innovations in misinformation and AI security, including:
o Intent modeling for misinformation
o Explainable misinformation prediction
o Harmful prompt detection
o Cross-lingual misinformation detection
o Retrieval-steering attacks
o Concept-driven prompting
Publications
Conference Papers
- WWW 2026 – Full research paper on intent modelling for misinformation detection presented at The ACM Web Conference (WWW) 2026.
- EMNLP – Two full research papers have been accepted 2026
Events & Engagements
Workshops Organized
- DHOW-MiLLA 2026 (Dubai, UAE)
Co-organized the Joint Workshop on Diffusion of Harmful Content on Online Web and Countering Misinformation in the Age of LLMs and Agents at The ACM Web Conference (WWW) 2026.
Summits & Forums
- AI Impact Summit 2026
The SS team participated and presented in the AI Impact Summit, engaging with researchers, industry leaders, policymakers, and practitioners.
Industry-Academia Collaborations & Visits
- IISC Bangalore
- IIT Bombay
- IIIT Hyderabad
- IIT Delhi
International collaborations -
Frontria is Fujitsu’s global consortium initiative focused on addressing emerging challenges in AI trust, AI security, misinformation, disinformation, and the safety of AI systems by bringing together industry, academia, and technology leaders to co-create practical solutions for a trusted digital society. A key contributor to this vision is Fujitsu Research India’s Security Sciences Business Unit, the unit focus on research spans trustworthy and secure AI, LLM safety, adversarial robustness, AI alignment, misinformation detection, synthetic media detection, and explainable AI. Through its work on building secure, resilient, transparent, and trustworthy AI systems, the Security Sciences BU directly supports Frontria’s mission of enhancing information reliability, combating AI-driven threats, and enabling the responsible deployment of next-generation AI technologies.
Meet our leadership
Dr. Rahul Mishra
Rahul Mishra is Research Scientist Manager and Head of the Security Science Lab at Fujitsu Research India, where he leads research to build AI systems that are not only capable but also secure, interpretable, and trustworthy for real-world deployment. With over a decade of experience in AI and NLP, his work spans trustworthy and secure AI, large language models, misinformation, explainability, and graph representation learning, with a current focus on LLM safety, adversarial robustness, retrieval-augmented generation, AI alignment, and the detection and prevention of misinformation and synthetic media. Prior to Fujitsu, he held research and academic positions at the University of Oslo, IIIT Hyderabad, the University of Geneva, and the University of Stavanger, and worked in industrial research at Samsung Research, IBM Research, and TCS Innovation Labs, including a visiting researcher role at ETH Zurich.
