The Executive Office has opened consultation on Northern Ireland’s first artificial-intelligence strategy, setting out a framework that treats public-sector AI as a governance, data, infrastructure, workforce, and service-design problem rather than simply another route to administrative automation.

The draft is organised around eight principles: human oversight, accountability and redress, data governance, technical safety and security, fairness and transparency, sustainability, societal benefit, and training and literacy. Consultation opened on 12 August and runs until 5pm on 7 October, after which responses will be analysed before a final strategy is brought to the Northern Ireland Executive for agreement.

Although centred on public-sector adoption, the strategy extends beyond rules for individual AI tools. It covers Northern Ireland’s wider AI landscape, governance, infrastructure and data, skills development, public-service transformation, and citizen trust, reflecting the reality that departments cannot obtain useful AI simply by procuring a model and attaching it to existing processes.

The consultation follows Northern Ireland’s 2025 Programme for Government commitment to establish an AI unit aimed at improving public-sector efficiency and effectiveness. The draft strategy provides the wider framework around that ambition, including how systems should be supervised, what data they can rely on, what capabilities staff need, and how people affected by AI-supported decisions should be treated.

Human oversight becomes an operating requirement

Human oversight sits first among the eight principles, while the strategy emphasises augmenting human capabilities rather than replacing human agency and decision-making. The practical meaning of oversight will vary enormously between an employee using AI to summarise a document and a system influencing access to a public service.

Meaningful human involvement requires more than putting an employee nominally into the approval chain. Staff need enough information to understand what a system has produced, authority to challenge or overturn it, time to exercise judgement, and a record of how decisions were made.

Accountability and redress create a corresponding requirement for service users. When an AI-supported process contributes to an incorrect decision, the organisation needs to know who owns the outcome and how somebody can challenge it, particularly where decisions affect benefits, health, education, regulation, licensing, or other services that cannot simply be switched for a competitor.

The strategy also couples governance with data quality and security. AI systems depend on information drawn from existing departmental estates, which can include legacy databases, inconsistent records, fragmented ownership, and services built for purposes far removed from modern AI workloads.

Public-sector AI inherits public-sector infrastructure

Including infrastructure and data as a specific theme acknowledges that adoption cannot be separated from the technology environment underneath it. Departments may want systems that search internal knowledge, automate repetitive work, or assist service delivery, but those tools still require authorised access to reliable data and enough technical integration to fit existing operations.

The same pattern can already be seen elsewhere in government. Ireland’s welfare department has been tying AI work to architecture, data, and governance contracts, illustrating how public-sector deployment often becomes an integration programme long before it resembles the frictionless demonstrations associated with general-purpose AI products.

Northern Ireland’s framework also includes technical safety and security, bringing cyber risk into the adoption model. AI can introduce new suppliers, data flows, interfaces, permissions, and dependencies, while generative systems add uncertainty around output quality and the information sent into models.

Sustainability is included as well, putting energy and resource use inside the governance framework rather than treating computing consumption as separate from technology choice. The consultation does not prescribe a measurement system, leaving later implementation to determine how departments balance sustainability against cost, performance, and service outcomes.

Skills will shape implementation

Training and literacy form another core principle because AI adoption changes the work expected of civil servants as well as the technology available to them. Employees need to know where systems are useful, where outputs require verification, what information can be entered, and when automated assistance is inappropriate.

The training challenge extends beyond ordinary users. Procurement teams need to assess suppliers, technical specialists need to understand integration and security, managers need to redesign processes rather than attach AI to inefficient ones, and governance functions need enough technical understanding to scrutinise systems without reducing every discussion to abstract principles.

The Executive Office is seeking responses from businesses, SMEs, academics, trade unions, local government, community organisations, public-sector bodies, and members of the public. That breadth reflects the fact that some of the hardest questions around public AI concern workers and service users rather than the capabilities of the underlying models.

Job security is explicitly listed among the issues raised by adoption, alongside accountability, fairness, energy usage, transparency, and public trust. The strategy does not settle those questions in advance, and the consultation means the eight principles remain a proposed framework rather than completed operating policy.

Once responses close on 7 October, the difficult work will be translating broad principles into rules departments can use when procuring, deploying, monitoring, and retiring actual systems. Human oversight, fairness, security, and transparency are straightforward to endorse at strategy level; they become harder when a service is under pressure to reduce costs, accelerate decisions, or process more work with the same workforce.

The final measure will therefore be whether those principles appear in procurement specifications, data practices, staff training, service design, audit processes, and routes for redress once AI begins influencing ordinary public administration.