LondonMinisters’ ChatGPT logs are subject to public disclosure.
Britain’s Information Commissioner’s Office confirmed that AI prompts and outputs generated by public-body staff fall under the Freedom of Information Act.
Public bodies previously rejected AI-related requests as vexatious; that defense is now untenable, a data-law expert said.
Sources: New Scientist
WHAT FOI ACTUALLY IS
The UK Freedom of Information Act 2000 gives anyone — citizen or not — the right to request recorded information held by public bodies. The test is whether the information is *recorded* and *held*, not whether it was meant to be kept.
THE RECORDED-INFORMATION TRIGGER
FOI does not cover what officials think or say in person. The moment a thought is typed into a system that logs it — email, Slack, or a ChatGPT prompt — it becomes recorded information. AI chat logs are no different in law from a draft memo on a shared drive.
WHY 'VEXATIOUS' WAS THE OLD SHIELD
Section 14 of the Act lets public bodies refuse requests deemed vexatious — disproportionate, harassing, or lacking serious purpose. Departments leaned on this to deflect novel AI requests. The ICO's clarification removes that cover: a request for ministerial prompts is no more vexatious than one for ministerial emails.
THE PRECEDENT THAT MATTERS
In 2005, Tony Blair's government tried to exempt cabinet discussions; the Information Tribunal forced disclosure of Iraq war legal advice. FOI in Britain has consistently expanded outward — from formal minutes to informal email to WhatsApp messages (ruled disclosable in 2023). AI prompts are the next layer.
THE COMPARATIVE LANDSCAPE
Different democracies have drawn the AI-in-government line in very different places. The UK is now among the most disclosure-permissive; the US FOIA framework is still litigating whether prompts are 'agency records'; the EU AI Act focuses on risk classification rather than transparency.
THE BEHAVIORAL EFFECT
FOI's deepest impact is not the disclosures themselves but the chilling effect on what gets written down. Officials shift to verbal briefings, ephemeral apps, or personal devices. The same shift will likely happen with AI — sensitive prompts will migrate to consumer accounts the state cannot subpoena.