DPhil in Artificial Intelligence Security
Advancing the fundamental theory of machine learning as applied to security, including but not limited to: model, data, learning, cyber and and agentic security.
Full time
- Upcoming application deadline for 2027-28 entry:
- 12:00 midday UK time on Wednesday 6 January 2027
- Expected length
- 3-4 years
- Expected start date
- 4 October 2027
- English language level
- Standard level required
Application deadlines
Final application deadline to be considered for Oxford scholarships
Expected to be the final application deadline
After the final application deadline has passed, this course may remain open or reopen if places are available.
The Radcliffe Camera. University of Oxford Images / Whitaker Studio
About the course
Admission to the DPhil in Artificial Intelligence Security is restricted to UK nationals (including those holding UK nationality alongside other nationalities).
The DPhil in Artificial Intelligence Security is a doctoral degree, designed to enable students with advanced Artificial Intelligence (AI) skills to tackle the most important problems in the security of AI.
This DPhil is in partnership with the UK Government Laboratory for AI Security (LASR) whose mission is to mitigate AI security risks to support national security and economic growth.
Research in the security of AI is informed by, and in turn advances, the fields of computer science, engineering science, mathematics, physics, and statistics. You will be based in the department of your lead supervisor; typically one of the following five University of Oxford departments:
- Department of Computer Science
- Department of Engineering Sciences
- Mathematical Institute
- Department of Physics
- Department of Statistics
You will be admitted based on a research programme and supervisor agreed by the funders. Variation of supervisor or research topic must be agreed by the funders.
AI has proven successful in applications as diverse as creative arts, autonomous vehicles, computer networks, knowledge work, software engineering and highly regulated and controlled domains such as critical national infrastructure, law enforcement, and national defence. These practical successes have outpaced our fundamental understanding of the underlying AI systems.
Most modern AI models that people are now familiar with are extremely large complex functions that are trained with vast amounts of data. The design choices of AI systems form highly complex processes that are rife with security issues. The rapid deployment of AI models comes with relatively limited understanding of the security of AI in general. Extending the application of AI models to highly regulated and controlled domains such as critical national infrastructure, law enforcement, and national security requires greater understanding of the models and their vulnerabilities.
This doctoral course will aim to:
- advance the theory of AI models
- reveal and, where possible, correct vulnerabilities in these systems
- highlight how design choices impact these vulnerabilities
- grow the national capacity of experts in the field of AI Security
Course structure
Key components of the teaching include:
- Mandatory research and security briefings
- A weekly AI Reading Group session
In addition to formal teaching, student experience and training will be enriched by:
- Research software engineering training and guidance
- Summer research training schools, provided by external organisations
- Occasional workshops
- Additional skills training made available by the Doctoral Training Centre and by your supervisor's department
Any costs you incur attending mandatory workshops or research training schools will be covered.
Core components
You will attend briefings and and seminars via an AI reading group
Research areas
You will have the opportunity to undertake research within the specialised themes of this course.
Course details
Entry requirements
For entry in 2027-28