Making AI Safer by Making Research More Inclusive: The University of York’s SAINTS Initiative
Who decides whether an artificial intelligence system is safe, and whose perspectives inform that judgement? As part of Veredus’s sponsorship of the Outstanding Contribution to Equality, Diversity and Inclusion category at the THE Awards 2026, Professor Ibrahim Habli, Director of the UKRI Centre for Doctoral Training in Safe AI Systems (SAINTS), and Helen Poyer, Centre Manager, at the University of York share how the Centre is addressing structural barriers to doctoral study and building a research community as diverse as the society it serves. Acceptably Safe – But Acceptable to Whom? In AI safety, the question of whether a system is ‘acceptably safe’ is fundamental. Safety is never absolute, and judgements about what is acceptable inevitably involve assumptions about risk and whose interests are being considered. “That raises an uncomfortable question: acceptable to whom, and under what circumstances?” explain Habli and Poyer. This tension sits at the heart of SAINTS and its approach to equality, diversity and inclusion. “EDI is not a box-ticking exercise. It is a research imperative.” AI learns patterns from the data it is trained on and from the metrics, nudges and models researchers choose. Many of these choices are socially and ethically sensitive, meaning that researchers’ backgrounds, disciplines, stories, experiences, preferences and constraints all matter. Yet the current AI safety landscape suffers from poor representation across many characteristics, particularly ethnicity and gender. For SAINTS, the vision from the outset has been to build a research community as diverse as the society it serves, recognising that the people designing and assuring AI systems should reflect those affected by them. Addressing Structural Barriers to Doctoral Study To realise this vision, SAINTS set out to address several persistent structural barriers in access to doctoral study and AI research careers. Gender underrepresentation remains a significant challenge. Higher Education Statistics Agency data cited by the Centre shows that in 2024/25, only 29% of postgraduate students enrolling in computer science identified as female. There are also distinct racial inequalities in doctoral recruitment. Analysis of doctoral applications at the University of York revealed that ethnic minority candidates faced disproportionately lower offer rates compared to White British applicants. Furthermore, traditional academic recruitment can disproportionately penalise non-traditional applicants, such as those transitioning from industry or non-technical backgrounds. Interdisciplinary barriers present another challenge. Conventional AI research often excludes vital perspectives from the humanities, law and health sciences, despite their importance to understanding the wider implications of AI safety. Removing Barriers Through Inclusive Recruitment SAINTS has responded through a targeted, multifaceted approach designed to address these barriers directly. Firstly, the Centre has implemented anonymised, fully blind recruitment where applicants’ names, pronouns, and previous institutions are redacted so candidates remain anonymous until the interview stage, mitigating unconscious bias. Secondly, to address racial underrepresentation, SAINTS ring-fences at least one doctoral place annually for a Black British candidate who meets its academic quality threshold. Thirdly, the Centre offers targeted outreach by hosting dedicated pre-application webinars to demystify the application process and appeal to non-traditional applicants. Diverse Leadership and Interdisciplinary Research The Centre leads by example and has embedded inclusion into its leadership structures. Its leadership team maintains at least 40% female representation and draws expertise from diverse domains, including active legal practitioners, such as part-time judges, and health policy advisers, including NICE policy experts. SAINTS also provides interdisciplinary and industry co-supervision, tailoring supervisory teams around individual student research topics across multiple university departments and external industry partners. Early Evidence of Progress The Centre’s EDI interventions have so far shown pleasing results. SAINTS has achieved a strong gender balance: in its first cohort of 11 doctoral researchers, 5 places were awarded to female students, and in the second cohort, 8 out of 13 places were awarded to female candidates. Additionally, by supporting early-stage applicants, the Centre has successfully established diversified talent pathways, recruiting candidates transitioning directly from industry routes rather than relying solely on standard academic pipelines. Inclusion as a Research Imperative For other higher education institutions seeking to make doctoral education more accessible, SAINTS offers five key insights: Ultimately, SAINTS’s approach reflects a fundamental principle: “The people designing and assuring these systems should reflect the people affected by them.” For Habli and Poyer, building a diverse research community is not simply a matter of widening participation, but a research imperative for developing safer AI systems.