Lucy Rigby, the newly appointed chief secretary to the Treasury, has delivered a blunt warning that the UK risks “choosing decline” if it fails to adopt artificial intelligence across public services. Speaking as Whitehall moves from early experimentation toward policy and procurement decisions, Rigby’s message is less about hype and more about competitiveness: governments that refuse to modernise with AI, she argues, will fall behind in productivity, service quality and administrative capacity—at precisely the moment when budgets are tight and demand for public support continues to rise.
Her comments land at a time when AI is no longer confined to pilots or academic discussions. Across government, the conversation has shifted from whether AI can be useful to how it should be governed, funded and deployed responsibly. That shift matters because public services are not like private-sector operations where experimentation can be contained to a single product line. In the UK, AI touches benefits, healthcare pathways, policing, local authority planning, immigration processes, education support and a wide range of citizen-facing services. The stakes are therefore both operational and political: efficiency gains must be weighed against risks to fairness, transparency, security and accountability.
Rigby’s framing—AI as a tool for operating “more efficiently and effectively”—is also a signal that the Treasury is trying to shape the direction of travel rather than simply react to it. The Treasury’s role in setting spending priorities and evaluating value for money gives it leverage over whether AI becomes a patchwork of isolated projects or a coordinated programme with measurable outcomes. In other words, the question is not only whether AI should be used, but whether it can be used in a way that produces demonstrable returns for taxpayers.
What “choosing decline” really means in practice
The phrase “choosing decline” is deliberately provocative, but it points to a concrete set of concerns. Public services are already under pressure from multiple directions: rising costs, workforce constraints, increasing complexity in casework, and citizens’ expectations shaped by digital experiences outside government. Many departments have spent years digitising forms, automating routine tasks and improving online access. Yet even with those efforts, backlogs persist and staff remain stretched.
AI enters this landscape as a potential multiplier. It can help triage demand, draft responses, summarise documents, assist with knowledge retrieval, and support decision-making workflows—especially in environments where large volumes of text and data are involved. For example, many government processes rely on interpreting complex guidance, extracting relevant details from case files, and producing consistent communications. AI systems can, in principle, reduce the time spent on these tasks and help staff focus on higher-value work such as judgement, investigation and human interaction.
But the “decline” argument is not just about speed. It is also about capability. If the UK does not build internal expertise—technical, legal, procurement and operational—it may become dependent on external vendors without the ability to steer outcomes. That dependency can be expensive, and it can limit the ability to adapt systems when policies change or when new risks emerge. In the long run, refusing to adopt AI could mean losing institutional learning: the government’s ability to understand what works, what fails, and how to improve systems over time.
Rigby’s warning therefore reads as a call for adoption with discipline. The alternative is not necessarily “no AI,” but slower adoption, fragmented adoption, or adoption that happens without the Treasury’s insistence on value for money and measurable performance. In that scenario, the UK might still use AI—just later, less coherently, and at higher cost.
From pilots to policy: the governance question becomes central
As AI moves from pilots to policy discussions, the debate increasingly focuses on how governments balance innovation with responsibility. Rigby’s remarks reflect that reality. She is not arguing for blind deployment; she is arguing for a rollout approach that treats AI as part of modern public administration rather than an optional experiment.
That distinction matters because public-sector AI is not a single technology. It includes different categories of tools: machine learning models trained on historical data; natural language processing systems that can interpret and generate text; computer vision systems that can analyse images; and decision-support tools that may recommend actions or highlight risk factors. Each category carries different risks and requires different controls.
In the UK context, governance questions typically include:
1) Accountability: Who is responsible when an AI system makes an error?
2) Transparency: How can decisions be explained to citizens and oversight bodies?
3) Fairness: Does the system perform consistently across different groups and circumstances?
4) Data protection and security: What data is used, where it is stored, and how it is protected?
5) Procurement and vendor management: Can the government audit performance and ensure compliance?
6) Human oversight: Where must humans remain in the loop, and what does “oversight” actually mean?
Rigby’s emphasis on rolling out technology across Whitehall suggests that these issues cannot be solved department-by-department in isolation. A coordinated approach is likely needed to standardise risk assessments, procurement requirements, evaluation metrics and operational safeguards. Without that coordination, departments may develop incompatible systems, duplicate work, or struggle to enforce consistent standards.
The Treasury’s role: value for money as a forcing function
One reason Rigby’s position is significant is that the Treasury is often the gatekeeper for large-scale transformation programmes. AI rollouts can be expensive, and the benefits are sometimes difficult to quantify early on. If AI is treated as a novelty, it can become a budget sink. If it is treated as a productivity and service-quality programme with clear targets, it can become a lever for reform.
Rigby’s message implies that the Treasury wants AI to be evaluated through the lens of value for money and productivity. That means asking questions such as:
– Which processes are most suitable for automation or augmentation?
– What measurable outcomes will improve—case handling times, backlog reduction, accuracy, customer satisfaction, or staff retention?
– How will performance be monitored after deployment?
– What is the cost of failure, including reputational and legal risks?
– How will the government ensure that AI systems remain effective as policies and data change?
This is where the “roll out across Whitehall” element becomes more than a slogan. Whitehall departments share common functions—identity verification, document handling, call centre operations, policy drafting, compliance checks, and internal knowledge management. If AI capabilities can be shared or standardised, the government can reduce duplication and accelerate learning. Shared platforms also make it easier to apply consistent safeguards and to audit systems.
However, standardisation is not straightforward. Departments have different legal frameworks, different operational cultures, and different risk profiles. A national approach must therefore be flexible enough to accommodate variation while still enforcing baseline controls.
A unique take: AI adoption as administrative capacity, not just automation
A common misunderstanding about AI in government is that it is mainly about replacing tasks. In reality, the most valuable use cases may be about administrative capacity—helping institutions handle complexity at scale.
Many public-sector problems are not simply “slow” or “manual.” They are complex: they involve interpreting rules, reconciling inconsistent information, dealing with exceptions, and communicating clearly to people who may be under stress. AI can assist with parts of this complexity, particularly where the work involves large volumes of text, structured and unstructured data, and repeated patterns.
For instance, AI can support:
– Document triage: identifying which cases require urgent attention and which can follow standard pathways.
– Knowledge assistance: helping staff find relevant guidance quickly and consistently.
– Drafting and summarisation: producing first drafts of letters, reports or responses that humans review.
– Case file summarisation: extracting key facts and timelines from long documents.
– Compliance checks: flagging inconsistencies or missing information for human verification.
In each of these examples, the goal is not to remove human judgement but to reduce friction and cognitive load. That can improve both speed and quality—if the system is designed and tested properly.
Rigby’s “choosing decline” warning can therefore be interpreted as a warning about capacity. If the UK does not invest in AI-enabled administrative capacity, it may struggle to meet demand with existing staffing levels. The result could be longer waits, more backlogs, and a decline in service quality—outcomes that citizens experience directly.
The risk side: why responsible use is not optional
At the same time, the public sector cannot treat AI as a purely technical upgrade. Responsible use is not a compliance checkbox; it is essential to maintaining trust and preventing harm.
AI systems can fail in ways that are subtle. They may produce plausible-sounding but incorrect outputs. They may behave differently depending on how prompts are phrased or what context is provided. They may reflect biases present in training data. They may also create new security vulnerabilities if not properly controlled—especially when systems are connected to internal networks or when sensitive data is processed.
In government, these risks are amplified by the fact that AI outputs can influence decisions that affect rights, access to services, and legal outcomes. Even when AI is used only as a drafting tool or a summariser, errors can propagate into official communications. That is why governance frameworks typically emphasise:
– Robust testing before deployment, including stress tests and scenario-based evaluations.
– Clear documentation of model limitations and intended use.
– Monitoring for drift and degradation over time.
– Strong access controls and audit trails.
– Human review requirements for high-impact decisions.
– Mechanisms for complaints and correction when AI contributes to an error.
Rigby’s comments do not dismiss these concerns; they implicitly challenge the idea that caution must always mean delay. The policy question is how to move forward while building safeguards that are proportionate to risk.
The political economy of AI: procurement, skills and public trust
Another dimension behind the push to roll out AI across Whitehall is the political economy of technology adoption. Governments face a choice between building capability internally and relying heavily on external suppliers. Both approaches have trade-offs.
If the UK relies too much on vendors without developing internal expertise, it may struggle to negotiate favourable terms