Tony Blair has a plan to rebuild the state around AI
Key Points
- Tony Blair Institute paper says state institutions are "simply not equipped" for AI
- Blair says AI will transform the world within a decade, faster than the Industrial Revolution
- OECD survey cited shows only 40% trust their national government
- Plan calls for shared digital identity, data standards and institutional memory across government
- Bounded AI agents with named human owners would chase commitments and trigger workflows
Former UK prime minister Tony Blair has told political leaders that government institutions are “simply not equipped” for AI and must be redesigned around it within a decade.
The warning comes in a new paper from the Tony Blair Institute for Global Change titled Leading in the Age of AI: How to Build an AI-Enabled State, published on Monday (21 September) with a foreword by Blair as its executive chairman.
“The question leaders should be asking themselves is not whether AI is a good thing or a bad thing. Rather, it is simply the thing: the defining technological reality of our age,” Blair wrote in the foreword.
He compared the shift to the Industrial Revolution, but said that comparison understates what is coming because AI will transform the world within a decade rather than over generations.
The institute cited the Organisation for Economic Co-operation and Development’s latest cross-national survey, which found only 40% of respondents had high or moderately high trust in their national government, while 43% reported low or no trust.
The paper argued that this gap between political ambition and visible results is feeding a wider crisis of confidence in government that AI-enabled institutions could help close.
Blair said AI productivity gains could help developed countries break the cycle of rising taxes and spending that often fails to improve public services, while developing countries could use it to expand access to scarce expertise.
He also listed the risks, including labour market disruption, the concentration of economic power, cheaper autonomous cyber-attacks and the potential for AI to enable new biological threats.
The institute’s proposal is a model it calls the Reimagined State, which uses AI to move knowledge, coordination and decisions across departmental boundaries rather than keeping them locked inside individual institutions.
The paper set out a three-part framework for getting there, covering institutional reform, digital foundations and the delivery of AI across government.
What the institute wants governments to build
The first part asks leaders to concentrate political attention on a small number of outcomes and to make it easier to create, adapt and wind down government units around those priorities.
The second calls for shared digital foundations, including digital identity, secure data exchange and common standards, so information can move lawfully between departments without bespoke negotiations each time.
The institute also wants governments to build a shared institutional memory that records significant decisions, the evidence behind them and the assumptions that informed them, so knowledge survives when ministers move or teams are reorganised.
It said AI-generated summaries must sit as a derived layer on top of official records and should never silently overwrite them, with every material synthesis identifying the specific evidence it draws on.
The third part covers deployment, starting with a common route for AI projects to move from pilots to system-wide adoption, with shared procurement and evaluation standards so departments stop buying the same technology separately.
The institute said contracts should require interoperability, data and model portability, audit rights and exit provisions so a model or supplier can be swapped without rebuilding the application around it.
It proposed an AI-enabled strategic intelligence and simulation capability at the centre of government, combining AI with economic modelling, forecasting and operational data to test how major decisions interact across the state.
The paper also recommended bounded AI agents that can track commitments, request updates and initiate approved workflows between formal interventions.
Each agent would need a named human owner, with authority that remains explicitly delegated, bounded, auditable and revocable, and the institute said agents cannot acquire political mandates.
“AI can identify a conflict between government objectives, but it cannot decide which should prevail. That requires political leadership,” the institute said.
The paper pointed to South Korea’s Economic Planning Board in the 1960s, Singapore’s Housing & Development Board and Economic Development Board, and Estonia’s digital state after 1991 as examples of leaders building new institutions to deliver national transformation.
It said the model is not designed to concentrate unconstrained power at the centre, and that greater state capability must be matched by greater transparency and accountability.
Systems making consequential decisions about people would need to generate logs showing the authority they operated under, the sources they used, the model and policy versions involved, and where human judgement was exercised.
“If leaders don’t guide this revolution, it will happen to them,” the institute said in a post on X.
