Safe Artificial Intelligence PhD at York: Lifelong Safety Assurance for Autonomous AI

Safe Artificial Intelligence PhD at York: Lifelong Safety Assurance for Autonomous AI

Quick Decision

This is a four-year, funded doctoral place in a multidisciplinary cohort at the University of York, not a vacancy with one pre-written thesis. Its research remit is lifelong assurance of AI-enabled autonomous systems and the safety implications of transferring decisions from people to increasingly autonomous systems. Research topics and two or more supervisors are matched after offers, taking account of a researcher’s interests, the team’s theme and available data or case studies. The main attraction is the combination of technical and socio-technical safety research, a structured cohort, and partner exchanges; the main decision risk for International fee-status applicants is that the public page says annual tuition is paid but does not explain who covers any fee above the UKRI Home contribution.

Stage 1 closes at midday on 11 January 2027. Home fee-status researchers are listed for a £30,000 tax-free 2027/28 stipend through TechFirst. The advertised International baseline is £21,805 for 2026/27; the 2027/28 amount is not yet confirmed, and International places are limited. Before investing in an application, an International applicant should obtain written confirmation of fee-gap coverage, award eligibility and the stipend applying to the offer. The initial application is an anonymised expression of interest, not a full proposal or supervisor-specific application. 2027–28 studentship details, application timeline and application guide set the controlling details.

Research Problem and Doctoral Work

What is unsolved and why it matters

SAINTS frames the problem as keeping AI-enabled systems acceptably safe as models, users, deployment settings, and operating conditions change. A one-off pre-deployment test or a model-level performance score may not show whether a system remains safe after it is adapted, integrated into a wider workflow, or used in a new context. The programme also treats safety as more than a property of code: human control, accountability, regulation, ethical acceptability, organisational practice, and impacts on people can matter to the assurance case. Its two umbrella themes are (1) lifelong safety from design through post-deployment and (2) safety as decision-making shifts from humans to AI systems. The SAINTS research themes and worked use cases show how these questions could arise in autonomous transport, ships, drones, industrial systems, healthcare and other contexts. These examples illustrate the programme’s scope; they are not advertised thesis assignments.

Work, methods and potential outputs

The eventual project is developed with the supervisors after an offer. The programme says it will reflect the researcher’s skills and interests, the team’s approach to a theme, and available datasets or case studies. Applicants therefore enter with a well-reasoned area of interest rather than a fixed research proposal or a promised dataset. Two or more aligned supervisors are assigned after acceptance; applicants are not expected to contact a preferred supervisor during Stage 1. The SAINTS FAQ and application guide are explicit on this point.

York’s recent research gives a useful, evidence-backed sense of the methodological neighbourhood, without predicting an individual student’s topic:

• Lifecycle safety assurance for machine learning. Paterson and colleagues’ 2025 AMLAS paper describes how safety assurance can be integrated into the development of ML components and how evidence can support a safety case for the larger autonomous system. Its object-detector example helps explain the move from model testing to system-level justification; it does not claim that an assurance case proves a system safe in every context. White Rose research record.
• Assurance across iterative engineering. Hodge and Osborne’s 2025 AgileAMLAS paper combines iterative software/ML development with safety-engineering activities and continuing assurance evidence. It responds to a mismatch between adaptive ML development and more sequential safety lifecycles, drawing on autonomous-robot projects. It is a process methodology, not a guarantee that every deployment risk can be eliminated. White Rose research record.
• Evaluation under unfamiliar conditions. Haider and colleagues’ 2024 experiments compare out-of-distribution detection approaches for reinforcement-learning systems. They show that the definition of “unknown” and the evaluation design can change whether a detector appears to distinguish harmless from potentially unsafe situations. This is a warning against treating a benchmark result as a complete safety argument. White Rose research record.
• Deployment and workflow context. Jia and colleagues’ 2025 healthcare case maps the clinical workflow, analyses hazards and risk, and constructs a safety case for deployment and ongoing monitoring of an AI-assisted diagnostic tool. It demonstrates why deployment arrangements can create risks not settled by product approval; its healthcare context is an example of socio-technical analysis, not a commitment to a medical-AI thesis. York Research Database record.
• Ethical assurance. Porter and colleagues’ PRAISE paper develops a structured argument pattern applying principles such as justice, beneficence, non-maleficence and human autonomy to a defined AI/autonomous-system context. The hypothetical use case shows how ethical claims can be made explicit and challenged alongside technical arguments. White Rose research record.

Taken together, these works show an active research environment spanning safety cases, ML development, evaluation, deployment and ethical reasoning. They also expose the central research challenge: evidence must be relevant to a defined system and operating context, and be updated as that context changes. The programme does not state which tools, formal methods, datasets, model types, compute provision or application domain an incoming student will use. Those depend on later topic matching.

School, Programme and Field Context

The degree is a University of York PhD hosted through the UKRI AI Centre for Doctoral Training in Safe Artificial Intelligence Systems (SAINTS CDT); public materials do not describe a joint or dual award. The programme draws on Computer Science, Law, Philosophy, Health Sciences and Sociology, with additional relevant disciplines listed in its application materials, including engineering, mathematics and economics. It is designed for a cohort of approximately 12–16 researchers, with a September 2027 start for this intake. Full-time study is four years; part-time study is eight years and is available only to researchers who do not need a Student visa. It is not an integrated Master’s programme. York’s programme page and the SAINTS FAQ describe the institutional and training arrangement.

The first six months place substantial emphasis on taught courses and team work; later study shifts more towards team research and the individual thesis while taught skills continue. The required training includes blocks in AI, Safe AI, and Law, Ethics and Society, seminars, quarterly workshops, and a normally mandatory secondment in Year 2 or 3 of at least three months full-time equivalent. The secondment is intended to connect the research to a partner or other external organisation and a real-world use case. Its host, timing and fit with an individual project are not fixed in the general information. SAINTS training details.

York’s Computer Science subject context is stronger and more directly relevant than its overall ranking alone. York reports that Computer Science was a UK top-ten department in the Times Higher Education analysis of REF 2021, with 100% of research impact and research environment rated at least 3*. The REF 2021 Computer Science and Informatics profile was 64% 4*, 34% 3* and 2% 2* overall. These are department-level research assessment results, not a measure of this particular CDT or a forecast of a student’s result. Separately, York is joint 158th overall in the QS World University Rankings 2027; that is an institution-wide measure, not a subject ranking. York REF results and institutional rankings.

The University’s Institute for Safe Autonomy (ISA) has a £15 million purpose-built facility with indoor and outdoor autonomous-system testing, robotics, advanced communications and quantum laboratories, among other resources. This is relevant context for an autonomy-focused programme, but public materials do not promise that a particular SAINTS topic will use the ISA laboratories or any named equipment. The Centre for Assuring Autonomy describes research around AI assurance, system design and governance, and Professor Ibrahim Habli’s York profile identifies him as Professor of Safety-Critical Systems and SAINTS director. These sources establish programme leadership and a research community, not an applicant-specific supervisory assignment. ISA facilities, Centre for Assuring Autonomy research and Professor Habli’s profile.

Supervisor and Lab Assessment

Research directions

The clearest supervision signal is collective rather than person-specific: York combines safety-critical systems and machine learning with law, philosophy, health, social sciences and operational partner settings. Professor Habli’s profile describes work on design and assurance of software-intensive systems, especially AI and autonomous applications, and interdisciplinary collaboration with ethicists, lawyers, social scientists and clinicians. The public programme says two or more supervisors will be matched after offers. It would therefore be premature to treat the programme director—or a paper author—as the applicant’s confirmed supervisor.

Lab, resources and collaboration environment

SAINTS lists more than 30 partner organisations, including companies, government organisations, regulators and charities. The FAQ says partners may contribute research exchanges, real-world data and case studies, guest teaching and facility access. The programme’s training page says a secondment is normally mandatory. These are substantive links to practice and governance, while the exact partner, dataset, access terms and confidentiality requirements remain project-specific. The ISA can support physical autonomous-system research where relevant, but the programme also accommodates non-laboratory work in law, philosophy, social sciences and health. Ask whether the matched project has a partner, data agreement, secondment host or equipment needs before accepting an offer.

Funding, Application and Living Cost

Stipend, tuition and fee status

For 2027/28, SAINTS states a £30,000 annual tax-free stipend for Home fee-status students under TechFirst, paid in instalments and normally increased with inflation. The page says annual tuition fees are paid and provides support for research events and conferences. International fee-status applicants are welcome, but the quoted £21,805 is the UKRI minimum for 2026/27; the 2027/28 minimum will be confirmed in early 2027. UKRI says international applicants can apply but places are limited. SAINTS funding, TechFirst details and UKRI studentship guidance.

There is a material ambiguity for International fee-status candidates. The SAINTS wording “annual tuition fees will be paid” does not specify whether the award covers the full International tuition rate. UKRI’s current guidance says the grant contributes up to the Home fee, and the student may need to contribute the difference. The training-grant terms also limit use of UKRI funds for that difference, leaving it to another source or an agreed waiver/arrangement. York’s 2027/28 Home research fee is £5,238 per year. Its indexed International research fee schedule lists £30,950 for PhD Safe AI, which would imply a nominal £25,712 annual difference if those rates apply; however, the extracted fee page retains an older 2026/27 heading, and neither that fee comparison nor UKRI policy identifies the terms of this specific award. Do not assume the candidate must pay the difference, and do not assume it is waived: ask SAINTS and York Admissions for written confirmation of the fee status, any waiver/co-funder, what the studentship pays each year, and how the four-year period is covered. UKRI student funding guidance, UKRI training-grant conditions, York Home fee schedule and York International research fee schedule.

York costs and buying power

The University’s 2026/27 postgraduate accommodation prices provide a concrete, but partial, budget reference. A 51-week postgraduate room is listed at £186 per week with a shared bathroom (£9,486/year), £206 per week ensuite (£10,506/year), or £224 per week for a large ensuite (£11,424/year); bills are included in university accommodation. These rents alone equal roughly 44–52% of the £21,805 2026/27 UKRI stipend baseline, before food, transport, personal costs and visa expenses. This comparison mixes the current accommodation year with a stipend baseline and should not be treated as a 2027/28 budget. Home students on the £30,000 TechFirst rate have a wider nominal margin, but their actual costs still depend on housing and personal circumstances. York advises applicants to budget for changing and individual costs. Postgraduate accommodation prices and York living-cost guidance.

UKRI does not pay visa fees or the Immigration Health Surcharge. York’s general fee guidance says research training support charges can apply to some projects, although the public page does not establish whether this programme or a particular student will incur one. Clarify these items, relocation costs and any stipend eligibility condition with the provider. UKRI support guidance and York Home research fee information.

Application stages and fixed dates

Stage 1 closes 11 January 2027 at midday. The official timeline lists decisions in the week beginning 8 February; in-person assessment days on 23 and 24 February; an online day on 25 February for international applicants; and final decisions in the week beginning 8 March. Official timeline.

The first stage asks for an online application form and an Expression of Interest containing:

• a research statement of no more than 1,000 words, aligned with at least one of SAINTS’ two themes and explaining novelty, significance, feasibility, multidisciplinary need and potential partner engagement; and
• a personal statement of no more than 500 words covering academic background, relevant work or voluntary experience, reasons for choosing SAINTS and intended career direction.

The EOI is assessed anonymously and should be readable to a multidisciplinary audience. A full proposal, named supervisor and references are not required at the start; two academic references are requested later if an offer is made. Stage 2 is a group selection event with panel interview, in person for most applicants and online for international applicants. The guide states that submitted applications cannot be changed after final submission. Application and EOI guidance.

Applicant Fit and Preparation

Required background and skills

The usual academic threshold is a first-class honours degree (or equivalent), or a 2:1 honours degree plus a Master’s degree (or equivalent). Relevant professional experience or equivalent qualifications may be considered for applicants with a non-traditional path. Applicants do not need a Computer Science degree, but basic programming knowledge is expected. Prior practical experience creating AI-enabled systems is desirable, not essential. The programme spans computer science, engineering, mathematics, economics, health sciences, law, philosophy and sociology; the chosen background must support doctoral work in the applicant’s area. Official eligibility and application guide.

Suggested preparation

A competitive EOI should make a small number of research questions understandable beyond one discipline. Useful preparation is to:

1. Select one SAINTS theme and one plausible real-world context, then explain what “safe” means in that context and how relevant evidence could change through deployment.
2. Show a feasible way to connect the candidate’s discipline to other required perspectives, such as ML/system evidence with law, human factors, clinical practice or governance.
3. Give examples of research, programming, analysis, writing or collaboration skills and explain what each example demonstrates; distinguish the required programming baseline from the preferred practical AI experience.
4. Write the research statement as an initial interest and rationale, not a fixed thesis plan: the programme supports proposal development during Year 1 and assigns supervisors after offers.
5. Follow anonymity, word-count, formatting and final-submission instructions exactly.

The published assessment criteria focus on relevance, novelty, significance, feasibility, the candidate’s understanding of collaboration, contribution to the cohort, and evidence of relevant skills. This makes clear articulation and cross-disciplinary reasoning as important to the application as a narrow technical idea. EOI assessment criteria.

Field Outlook and Career Market

The field is broadening from pre-deployment model testing toward evaluation, assurance and governance over the life of systems. The UK AI Security Institute’s agenda describes work on measuring frontier-model capabilities, evaluating mitigations, controlling advanced systems and understanding risks from autonomy and societal deployment. The Government’s 2026 AI action-plan update treats safe and trusted AI development and assurance as part of wider AI adoption and capability policy. These are national research and policy signals, not evidence that SAINTS students will work with AISI or that there is a particular number of open AI-safety jobs. AISI research agenda and AI Opportunities Action Plan: One Year On.

TechFirst is likewise a broad government-backed doctoral talent-pipeline scheme. Its participation supports the programme’s stated link to UK technology capacity, but it does not establish a job entitlement or that any role is reserved for graduates. TechFirst Doctoral Support.

Potential routes suggested by the programme and general career evidence include academic research, AI safety or product assurance, compliance, policy, regulation, legislation and public-sector work. Prospects’ general AI career guide also describes research, policy/ethics, product and engineering roles across sectors. It is not a forecast for AI safety specifically. SAINTS says some partners expect to offer graduate employment opportunities, but neither the programme nor the general career sources guarantee that a place, interview or job will follow. The programme’s own FAQ notes that many AI careers do not require a PhD, so the case for this four-year commitment depends on the value of original research, structured training and specialist safety-assurance expertise to a researcher’s goals. SAINTS career FAQ and Prospects AI career overview.

Next Actions and Questions to Verify

Before applying

• Confirm Home or International fee status and whether the 2027/28 studentship pays the complete International fee, including any difference over the UKRI Home contribution. Ask for the amount, source and duration of any waiver or co-funding in writing.
• If applying as International, ask whether the £21,805 2026/27 baseline is the amount currently budgeted for the offer, when the 2027/28 figure will be set, and whether the offer will include any adjustment.
• Check the basic programming expectation against prior experience and ask whether any 2027-entry pre-course support is planned if preparation is below the expected level; do not rely on an older FAQ’s cohort-specific date.
• Prepare the anonymized 1,000-word research statement and 500-word personal statement; keep a long-term email address active and review all formatting and submission rules.
• Attend one of the 16 November or 8 December 2026 information webinars if useful; the dates and times are listed in the official timeline.

At interview or before accepting

• Ask how two or more supervisors will be matched, when the thesis topic is finalized, and how a proposed interest connects to the team, available data/case studies and partner needs.
• Confirm any secondment host, expected timing, travel/cost arrangements, confidentiality constraints and how the exchange fits the thesis.
• Ask which York laboratories, computing resources or external datasets are actually available for the matched project, and whether a research training support charge applies.
• Before accepting, obtain the formal funding schedule, fee status, stipend amount by year, tuition coverage, visa/IHS and relocation exclusions, and any project-specific charges.

Evidence and Sources

Project, programme and application

• SAINTS 2027–28 studentships
• University of York PhD in Safe AI
• SAINTS FAQs
• 2027–28 application timeline
• How to apply and EOI assessment
• SAINTS training programme

York research environment and academic work

• York Institute for Safe Autonomy facilities
• Centre for Assuring Autonomy research
• Professor Ibrahim Habli profile
• York QS and other institutional rankings
• York REF 2021 results
• Haider et al. 2024: OOD detection and RL safety
• Paterson et al. 2025: AMLAS safety assurance
• Hodge and Osborne 2025: AgileAMLAS
• Jia et al. 2025: deployment safety case for diagnostic AI
• Porter et al. 2024: PRAISE ethics assurance argument

Funding, costs, sector context

• SAINTS TechFirst funding details
• UK Government TechFirst Doctoral Support
• UKRI-funded studentships
• UKRI student financial support
• UKRI training-grant terms
• York Home research fees for 2027/28
• York International research fee schedule
• York living-cost guide
• York postgraduate accommodation prices
• UK AI Security Institute research agenda
• AI Opportunities Action Plan: One Year On
• Prospects: How to get a job in AI