Securing Generative AI Model: Detection of Advanced Backdoor Attacks
Investigate and develop novel detection and defense mechanisms against advanced backdoor attacks in generative AI ecosystems.
Best for:Best for computer science graduates with a strong interest in AI security, machine learning robustness, and generative models.
Overview
This PhD position at Université Paris-Saclay focuses on investigating and detecting backdoor threats within generative AI model ecosystems, covering standalone models, retrieval-augmented generation systems, and LLM-based agents. The researcher will develop novel detection and defense schemes against stealthy trigger-based attacks tailored for real-world deployment scenarios. Expected deliverables include defensive mechanisms, open-source code releases, and comprehensive evaluation benchmarks for the scientific community. The project is hosted at the Laboratoire Instrumentation Intelligente, Distribuée et Embarquée under the supervision of Aurélien Mayoue.
What you'll do
- Investigate vulnerabilities and backdoor attack vectors within open-source generative AI models and RAG systems.
- Design and implement novel detection algorithms to identify poisoned components and knowledge bases.
- Build robust evaluation benchmarks for real-world deployment scenarios of generative models.
- Release developed defense tools and codebases as open source for community use.
- Publish research findings in peer-reviewed conferences and scientific journals.
Requirements
Degree required: Master's degree or equivalent in Computer Science, Artificial Intelligence, or a related field
Degree offered: PhD
Language: French: None; English: None
Instruction: English
- Strong background in machine learning and deep learning principles
- Programming proficiency in Python and relevant AI/ML frameworks
- Familiarity with generative AI models and large language models
- Understanding of security vulnerabilities or adversarial robustness in machine learning
Funding & terms
How to apply
Deadline: 2026-10-31
Interested applicants should submit their application materials through the designated portal before the stated deadline. Ensure all required academic and identification documents are attached.
- CV
- Cover letter
- Academic transcripts
Contact
Create a free profile to unlock your personal match score, the full requirements, funding details and how to apply — in English or Persian.