Contact us

AI Security DPhil

AI systems are advancing faster than our understanding of how to secure them. The DPhil in AI Security brings together research in computer science, engineering, mathematics, physics and statistics to investigate how AI systems can be better understood, tested and protected.

Explore admissions View supervisor profiles

A DPhil focused on AI security

If you already have advanced AI skills and want to investigate fundamental questions about its security, this DPhil offers a research programme spanning computer science, engineering, mathematics, physics and statistics.

You will join the department of your lead supervisor and pursue research connected to the priorities of the Laboratory for AI Security Research (LASR).

The programme is run in partnership with the UK Government laboratory, whose mission is to mitigate AI security risks in support of national security and economic growth.

People discussing research around a table

Why research AI security?

AI systems are being used in increasingly sensitive settings. Understanding their security has not kept pace with their deployment.

Modern AI models are complex, often trained on vast quantities of data. The choices behind their design and training can create vulnerabilities that are difficult to identify or explain. These questions matter particularly in areas such as critical national infrastructure, law enforcement and defence.

The DPhil investigates the theory of AI models, where vulnerabilities arise and, where possible, how they can be addressed. It also examines the security consequences of decisions made when AI systems are designed and trained.

Studying for the DPhil

Your research is based in your lead supervisor’s department, with a weekly AI reading group, seminars and opportunities for further training alongside your project.

Explore the programme →Researchers working at computers
Seminars and training

The weekly AI reading group is a required part of the programme. Seminars bring together Oxford researchers and invited speakers, while occasional workshops and briefings address research and security practice.

You can also draw on research software engineering guidance and skills training through the Doctoral Training Centre and your supervisor’s department.

Research facilities and support

The programme provides a budget for a laptop or tablet, together with an annual fund to support research and travel.

High-performance computing facilities and research software engineering support are available, and you will have office space in your supervisor’s department.

Applications open: November 2026

Admissions

Applications open: November 2026

The first step is to contact a potential supervisor in your area of interest. You will need to agree a research proposal with them, which will be considered for alignment with LASR’s research priorities.

DEADLINE
Date to be confirmed
INTERVIEWS
Interview notification: typically late January or early February
START
Date to be confirmed
Find out more

Research themes

Your proposed project should align with at least one of LASR’s four foundational research priorities.

Model security

How model architecture and design choices affect security.

Data security

How to protect private training data and establish trust in generated media.

Learning security

How training processes affect information exposure and robustness to attack.

Frontier model security

How to evaluate and safeguard frontier models, including large language models.

Explore our research →

Our supervisors

See all supervisors