Research Title:
Adversarial Robustness in Multi-Modal Foundation Models
Funded by:
The Italian Institute of Artificial Intelligence (AI4I), in collaboration with Politecnico di Torino
Supervisors:
Nicola Franco, Luca Cagliero, Danilo Giordano
Contact:
nicola.franco@ai4i.it
The Italian Institute of Artificial Intelligence for Industry (AI4I), in collaboration with Politecnico di Torino, invites applications for a PhD position focused on Adversarial Robustness in Multi-Modal Foundation Models.
Multi-modal AI models that integrate vision, language, and audio processing are becoming increasingly prevalent in critical applications such as content moderation, customer service automation, and AI-assisted software development. However, these systems introduce novel attack surfaces arising from cross-modal interactions, where adversarial inputs in one modality can exploit semantic inconsistencies or vulnerabilities when processed jointly with other modalities. This research aims to systematically investigate these vulnerabilities and develop novel attack methods that expose weaknesses in cross-modal processing.
The Role
The PhD candidate will join the Database and Data Mining (DBDM) group at Politecnico di Torino, working in close collaboration with researchers at AI4I. The candidate will explore the intersection of machine learning security, multimodal reasoning, and adversarial robustness, contributing both theoretical insights and practical tools for secure multi-modal AI systems.
Research Objectives
- Investigate adversarial vulnerabilities arising from cross-modal interactions in large multi-modal models.
- Develop multi-modal adversarial attacks exploiting vulnerabilities across modalities.
RESEARCH PLAN
Attack Surface Mapping & Early Attacks
Analyze multi-modal model architectures to identify exploitable interaction points. Develop a taxonomy of attack vectors exploiting modality interplay and publish initial findings. Gain foundational knowledge through courses in adversarial machine learning, information theory, and multi-modal AI.
Dataset Development & Benchmarking
Create a dataset of synthetic injection attacks categorized by technique, vulnerability, and context. Conduct experimental evaluations to quantify vulnerabilities in real-world systems such as content analysis, conversational AI, and code assistants. Submit findings to top-tier venues in machine learning security and AI.
Scalable Attack Generation & Defense Design
Develop parameterized attack templates capable of automatically generating diverse multi-modal adversarial examples. Build and test open-source tools for attack simulation and robustness evaluation. Propose best-practice frameworks for designing secure and trustworthy multi-modal AI systems.
Required Qualifications
- Strong background in machine learning and artificial intelligence, with emphasis on deep learning architectures.
- Proficiency with deep learning frameworks such as PyTorch or TensorFlow.
- Desirable experience in multi-modal models, computer vision, natural language processing, or audio processing.
What We Offer
- Joint supervision between AI4I and Politecnico di Torino, fostering academic collaboration.
- Opportunity to contribute to open-source research, publications, and community datasets.
- Access to high-performance computing infrastructure and expert network.
- A collaborative, international research environment with exposure to both academic and industry partners.
Start Date:
01/03/2026
How to Apply
Applications for this position are managed by Politecnico di Torino.
Please apply via the university’s official PhD portal:
For More Information
ABOUT US
AI4I – THE ITALIAN RESEARCH INSTITUTE FOR ARTIFICIAL INTELLIGENCE FOR INDUSTRIAL IMPACT
AI4I has been founded to perform transformative, application-oriented research in Artificial Intelligence.
AI4I is set to engage and empower gifted, entrepreneurial, young researchers who commit to producing an impact at the intersection of science, innovation, and industrial transformation.
Highly competitive pay, bonus incentives, access to dedicated high-performance computing, state-of-the-art laboratories, industrial collaborations, and an ecosystem tailored to support the initiation and growth of startups stand out as some of the distinctive features of AI4I, bringing together people in a dynamic international environment.
AI4I is an Institute that aims to enhance scientific research, technological transfer, and, more generally, the innovation capacity of the Country, promoting its positive impact on industry, services, and public administration. To this end, the Institute contributes to creating a research and innovation infrastructure that employs artificial intelligence methods, with particular reference to manufacturing processes, within the framework of the Industry 4.0 process and its entire value chain. The Institute establishes relationships with similar entities and organizations in Italy and abroad, including Competence Centers and European Digital Innovation Hubs (EDIHs), so that the center may become an attractive place for researchers, companies, and start-ups.
