UK government must confront hard truths if it wants to lead in AI

The UK government has been clear about its ambition to position itself as a global leader in AI. But beneath the strategy papers and summit speeches lies a more uncomfortable truth – the state must first confront how it understands, uses and governs data.

What is considered basic data literacy in the commercial world is, in parts of government, still emerging, said Snowflake field CTO, Fawad Qureshi.

“There is a knowledge gap between the public and the commercial sector. What we assume is common sense or straightforward is not always so common in government,” said Qureshi.

That gap matters, because in the age of AI, data is the foundation on which policy, services and public trust are built.

Bias starts upstream

The idea that “data is neutral” is one of the most persistent myths, said Qureshi.

“We are all humans. We all have our own biases,” he explained. “Data is capturing those behaviours – it’s not neutral. The risk is that we amplify those biases and put people at systematic disadvantage.”

For government, this means investment must shift upstream. Before building AI models, departments need to prioritise data quality, provenance and diversity, and critically, context.

Qureshi points to work on refugee data as an example: “Why should someone fleeing across borders trust you enough to be honest in a survey? For them, you are an adversary. If the data is collected in a context of fear, it won’t reflect reality.”

If the UK wants to lead in AI, it must first lead in understanding its own data.

Trust is fragile – and easily broken

Research into public attitudes towards government data sharing consistently shows that public trust is conditional on transparency, accountability and clear public benefit.

The government’s own Public Attitudes to Data and AI Tracker Survey found that the NHS, for example, remains one of the UK’s most trusted institutions for handling data, but trust falls significantly where people perceive weaker oversight or commercial involvement.

Meanwhile, recent Department for Science, Innovation and Technology (DSIT) research into the proposed National Data Library found that citizens often default to concerns about “undisclosed intentions” unless there is clear governance, transparency and explanation around how data will be used.

“When you collect data for one purpose, use it for that purpose. Don’t quietly use it for something else. You will break trust – and once it’s gone, it’s gone,” warned Qureshi.

Designing for data protection, therefore, must go beyond compliance checklists. It requires thinking about long-term societal impact.

Scaling AI requires accountability, not just automation

AI promises efficiency at scale. But scaling decisions is not the same as scaling accountability.

“AI can automate decisions, but accountability cannot be automated,” said Qureshi, who believes that for government, that means ensuring decisions can be explained, challenged and owned.

There is also a perception challenge: “People are much more tolerant of human mistakes than machine mistakes. One failure in an automated system, and it becomes a national story,” he said.

Elsewhere, the rise of synthetic data and deepfakes is reshaping the information landscape.

“We are moving into a world where you can’t always tell what is real, so you need to build verification into the system itself,” said Qureshi.

For government, this means investing in provenance and trust frameworks alongside analytics, ensuring that data can be traced, validated and trusted.

The leadership challenge behind it all

These issues ultimately point to a deeper problem: leadership and capability.

“You cannot regulate what you do not understand,” said Qureshi, who argues that the current government model – constrained pay, heavy reliance on contractors and structural inefficiencies – makes it difficult to attract the talent needed to close that gap.

“When the best people are not in government, you end up outsourcing policy work you cannot fully understand.”

At the same time, global competition for AI talent is intensifying, he said.

“AI is the new moon race. If you want to attract the best talent, you need the right environment – not just strategy documents.”

Becoming an AI leader is not just about investing in models or infrastructure. It requires a systemic shift, said Qureshi.

Leaders should:

• Treat data as a strategic asset, not an afterthought
• Protect and nurture public trust as a core capability
• Design for long-term societal impact, not short-term compliance
• Embed accountability into every automated system
• Build verification into the fabric of digital services
• Create an environment where top talent wants to work in government

Said Qureshi: “Trust is earned in drops and lost in buckets. One misuse of data, and people remember it for decades.”