New rules confirm public has a right to see how UK government uses AI

Text, images and other content produced by UK government departments and other public bodies using artificial intelligence are subject to freedom of information (FOI) laws, regulators have confirmed – potentially opening the door for the public to gain access to ministers’ ChatGPT or other chatbot records.

The Information Commissioner’s Office (ICO), the UK’s data-protection agency, has released new guidance confirming that “If staff at a public authority use AI for work purposes, the information generated will be subject to FOIA [the Freedom of Information Act] along with the prompts used”.

Last year, New Scientist successfully requested the then-UK tech secretary Peter Kyle’s ChatGPT logs under FOI legislation, in what is believed to be a world first. That triggered subsequent requests from other news outlets to obtain other information, but many have either been rejected on cost grounds or labelled as “vexatious”, an umbrella term that allows authorities to reject a request.

The clarification by the ICO could change that. “It will be very difficult for public authorities now to claim that AI-related requests are not subject to FOIA,” says Jon Baines at Mishcon de Reya, a London-based law firm.

“The ICO guidance is generally sensible, and won’t come as a surprise to most practitioners,” he says. “If information is ‘held’ in a recorded form by a public authority, wherever, and on whatever system, it is subject in principle to FOIA – so that must also apply to inputs to and outputs of AI systems and tools.”

“I think it should be uncontroversial,” says Tim Turner, a data-protection expert based in Manchester, UK.  “If records are captured by a public servant doing their job, they’re in scope,” he says. “That should be true for AI interactions and Post-it notes alike.”

The new guidelines could potentially enable successful requests for the prompts used by government workers in AI tools. The ICO has also suggested that public bodies may be required to use AI to summarise large documents or datasets when responding to requests, allowing them to answer requests previously dismissed on cost grounds.

Some have criticised the ability to obtain AI chat logs using FOI legislation. Matt Clifford, the chair of the UK’s Advanced Research and Invention Agency (ARIA), recently posted on social media that the ruling to release Kyle’s interactions with ChatGPT was “absurd”, “hugely corrosive and more or less guarantees that no minister will (say they) use AI”. Unusually for a public body, ARIA is exempt from FOI laws.

When asked if the new guidelines were triggered by New Scientist’s successful FOI request, the ICO declined to answer. A spokesperson says: “We regularly attend events and seek feedback on areas where both public authorities and requestors would value additional clarity and guidance. Our recent guidance on AI and FOI reflects what we have been hearing from organisations, and we tested the content with our external stakeholders to make sure it was as clear and useful as possible.”