OpenAI and the UK Government: Eight Months After the Deal, No Pilots Have Started

The UK government has still not launched a single pilot with OpenAI technology, even though the partnership with the maker of ChatGPT was signed eight months ago and presented as an important step toward public service reform. A request under the Freedom of Information Act showed that the loud announcement has not yet been followed by real trials.

A partnership without pilots

When London announced the memorandum with OpenAI, it was not simply about getting acquainted with a new technology. Ministers presented the agreement as a tool that would help modernize government services and use artificial intelligence to tackle major societal challenges. Against the general excitement around generative AI, it looked like a signal: the British government wants to be among those putting such systems to work not in presentations, but in the day-to-day operations of its departments.

In the British system, such initiatives are usually tied to digital modernization: faster replies to citizen inquiries, automated document drafts, support for operators and analysts, and less pressure on the front office. That makes the absence of even a limited trial launch especially stark. The expectation was practical application, not a long stage of discussions and framework agreements.

But eight months later, the picture looks noticeably more modest. The freedom of information request found no evidence that the government had conducted any tests of OpenAI technology at all. In other words, a gap has opened between public rhetoric and actual action. What matters here is not the bureaucratic delay itself, but that the partnership was presented from the start as a practical step toward reform, not as a framework intention for the future.

The partnership was supposed to help "tackle society's biggest challenges."

Why the launch is stalling

The lack of pilots does not necessarily mean the collaboration has collapsed. In the public sector, even a limited test of a new AI system requires long preparation: defining use cases, agreeing on handling of personal data, checking the quality of the model's answers, embedding it in existing processes, and working out who is responsible for errors. Officials are especially cautious about tools that may affect citizen requests, documents and official decisions.

Before a real launch, several stages usually have to be completed:

• pick a specific task where AI saves time rather than creating extra risk
• check how the model handles sensitive and restricted data
• set rules for human review and limits on automation
• clear legal and procurement procedures
• fix metrics: processing speed, quality, cost and error count

There is also a purely political problem. Signing a memorandum is much easier than choosing a department, taking one real process, setting a KPI and publicly showing the result. Once a pilot begins, questions immediately arise about reliability, transparency, citizens' rights and who will answer for a failure. That is why government bodies often agree on a strategic partnership faster than they reach the stage where AI has to work on live cases.

The problem is that without these pilots, the most basic promises cannot be checked. Without tests, it is impossible to say honestly whether OpenAI will speed up the handling of inquiries, relieve staff, or simply add a new layer of control and manual verification. Against this backdrop, the story becomes a test of how mature government demand for generative AI is: businesses are already putting models into internal processes, while the state still has to prove it can move from announcement to execution.

What this means

The story of the British government and OpenAI is a reminder of a simple fact: a long operational distance lies between signing a partnership and truly deploying AI. For the market, it is a signal that even large states move from loud announcements to working use cases more slowly than it appears from the outside.