The growing adoption of generative artificial intelligence (gen AI) across Welsh public services presents opportunities to improve efficiency and accessibility, but also raises questions about regulation, accountability and the quality of AI-generated information, according to a new Senedd Research briefing.
The briefing, Generative AI – an overview, examines the technology’s potential applications alongside risks including inaccurate outputs, discrimination, privacy breaches and the impact on employment.
It also considers the emergence of AI agents, which can autonomously select tools and carry out tasks, and the additional security and accountability challenges associated with giving AI systems greater autonomy.
The research comes as Wales develops its approach to AI following the publication of the Welsh Government’s AI Plan in November 2025.
AI adoption across Welsh public services
The briefing identifies opportunities for generative AI to improve the efficiency, quality and accessibility of public services, including through document drafting, information summarisation, record updates and data analysis.
It reports that the Welsh Government allocated £1 million in October 2025 to support AI adoption across Welsh public services.
Projects are underway to evaluate AI scribe tools, deploy a platform that allows public sector organisations to test and use AI, develop chatbots and deliver guidance and training for public sector workers.
The Welsh Government is also working with the UK Government to assess how its Humphrey AI toolkit could be made available and used in Wales.
However, the briefing notes that evidence of AI’s wider economic impact remains at an early stage. Citing Stanford University’s 2026 AI Index Report, it says productivity gains are measurable in specific tasks, but evidence of economy-wide improvements remains mixed.
Welsh-language AI presents opportunities and challenges
The research identifies the Welsh language as an important consideration for public sector AI adoption. Gen AI could help public bodies improve bilingual service delivery, including through conversational tools that allow citizens to access services in Welsh.
However, the availability of high-quality training data remains a challenge. Generative AI models are typically trained on large quantities of internet data, much of which is concentrated in languages such as English.
Work is underway to address these limitations. University College London, NVIDIA and Bangor University have developed a large language model designed to support AI reasoning in Welsh and other UK languages, with potential applications in healthcare, education and legal services.
The Welsh Government has also worked with technology companies to improve Welsh-language outputs in tools including Microsoft Copilot and ChatGPT.
If you liked this content…
No trust, no problem? Why identity security is central to generative AI in government
Governments need to be mindful of shadow AI risk
Addressing the skills gap in gen AI and security
Q&A: Andreas Kollegger on combatting gen AI’s hallucinations
Five ways government can turn responsible AI principles into practice
AI agents raise further security concerns
The briefing examines the implications of AI agents, which can use multiple computational tools to complete tasks and automate workflows. While these systems could help public sector organisations automate administrative processes, their ability to act with greater autonomy introduces additional risks. In particular, the research highlights prompt injection attacks, in which malicious instructions are hidden within information processed by an AI system.
Such attacks can manipulate AI agents into taking unintended actions, potentially including sharing or deleting data where agents have been granted the necessary permissions.
The briefing also raises questions about accountability when AI systems operate without direct human intervention, highlighting the importance of meaningful human oversight.
Regulation and devolved responsibilities
The research notes that there is currently no single, comprehensive legislative framework specifically covering AI. Instead, the technology is regulated through existing legislation and sector-specific frameworks.
The Welsh Government’s November 2025 AI Plan did not propose additional legislation. It established an Office for AI and a Strategic AI Advisory Group to support the government’s approach.
The briefing also highlights the division of legislative responsibilities between Wales and Westminster.
The protection of personal information, intellectual property and internet services are reserved matters, limiting the Senedd’s ability to legislate independently in these areas.
However, existing Welsh legislation remains relevant to AI adoption. Under the Well-being of Future Generations (Wales) Act 2015, public bodies must consider the long-term impact of their decisions, alongside immediate benefits.
The briefing also identifies the Workforce Partnership Council’s guidance on ethical AI and algorithmic management as a relevant resource for organisations considering the implications of AI for workers.
Balancing innovation with public trust
The Welsh Government’s AI Plan is organised around four pillars: Economic Growth, Educating Wales, Equitable Delivery, and Excellence & Trust. It sets out an approach intended to support economic development and public service improvements while emphasising equity and collaborative decision-making.
The briefing contrasts this with the UK Government’s broader emphasis on using AI to drive economic growth and productivity.
It also highlights challenges facing policymakers as AI technologies evolve, including difficulties defining harm, establishing accountability and developing regulation that can respond to new capabilities.
