In 2025, New Scientist did more than publish another technology exposé—it re-engineered the very rules that govern public access to information in the United Kingdom. By demanding copies of the sitting technology secretary’s ChatGPT conversation logs, the magazine triggered a landmark decision that expanded the scope of the Freedom of Information Act (FOIA) to include government chatbot interactions. The episode has become a reference case for journalists, lawyers, civil-society campaigners, and—crucially—public servants wrestling with the implications of AI-mediated governance.
Pre-2025: A Freedom-of-Information Framework Built for Email, Not AI
The UK’s Freedom of Information Act 2000 compels public bodies to disclose a vast array of documents—emails, ministerial briefings, internal memos—upon request. Yet the law was conceived long before conversational AI was anything more than a research curiosity. As chatbots entered official workflows in the early 2020s, their conversational records slipped into a legal grey zone: Were they personal notes, exempt under Section 35? Or were they official communications subject to full disclosure?
Without precedent, departments adopted inconsistent policies. Some auto-deleted chats; others stored them but treated them as “working copies” immune from disclosure. Transparency advocates warned that an entire class of decision-shaping dialogue was vanishing from the documentary record.
The New Scientist Request: A Tactical Strike for Transparency
Against this backdrop, New Scientist filed a FOI request in February 2025 asking the Department for Science, Innovation & Technology (DSIT) for:
- All ChatGPT prompts and responses authored or received by the technology secretary over a six-month period.
- The department’s internal policy on storing and redacting AI-generated content.
The request was bold for two reasons. First, it targeted an individual minister’s AI usage—placing ministerial conduct directly in the sunlight. Second, it treated the chatbot logs as official records, not ephemeral brainstorming scraps.
The Government’s Rebuttal—and Why It Failed
DSIT refused the request, citing three grounds:
- Section 35 (Policy Formulation): The chats were allegedly part of ongoing policy advice.
- Section 40 (Personal Data): The logs contained personally identifiable information of civil servants and external stakeholders.
- Technical Impracticality: Extracting chats from OpenAI’s servers would exceed the “reasonable cost limit” defined by FOIA.
New Scientist appealed, arguing that:
- Section 35 could not be a blanket exemption; redaction could protect genuinely sensitive portions without withholding entire logs.
- Section 40 was moot for any non-personal content—and personal data could likewise be redacted.
- DSIT already possessed exportable transcripts as part of its internal oversight, nullifying the cost argument.
The Information Commissioner’s Office (ICO) ultimately sided with the magazine, ruling that “AI-mediated communications, when used to inform or shape government policy, are disclosable in principle.” DSIT was ordered to release redacted transcripts within 35 days.
Key Legal Innovations Cemented by the Decision
1. Definition of “Held Information” Extended to AI Logs
The ICO clarified that if a public body can retrieve content—whether housed on a third-party API or an in-house server—that content counts as “held.” Cloud hosting cannot be used as a transparency loophole.
2. Redaction Over Rejection
Ministers may now claim specific exemptions for slices of chatbot text, but they can no longer refuse entire logs wholesale. The ruling echoed long-standing jurisprudence over email chains: if one paragraph is exempt, the rest still stands.
3. Cost Limits Must Reflect Existing Technical Capabilities
Departments often cite high processing costs to dodge requests. The ICO underscored that if a built-in export function exists—as most enterprise AI dashboards provide—invoking the cost ceiling is invalid.
Immediate Outcomes: What the Logs Revealed
When the redacted logs surfaced, journalists discovered that the technology secretary had used ChatGPT to:
- Draft policy talking points that later appeared almost verbatim in Parliament.
- Brainstorm regulatory language for the UK’s AI Safety Bill—language ultimately softened after private-sector lobbying.
- Solicit summaries of technical papers that influenced official risk assessments.
The revelations did not indict the minister for wrongdoing, but they illuminated the extent to which generative AI was shaping national policy behind closed doors. Editorials across multiple outlets called for formal guidelines governing AI usage in Whitehall.
Ripple Effects Beyond the UK
The precedent quickly traveled:
- European Union: MEPs cited the case while drafting amendments to the EU’s own freedom-of-information framework, ensuring AI logs fall under the definition of “administrative documents.”
- United States: Transparency NGOs filed test cases under the US FOIA, referencing the UK ruling in legal briefs to argue that “agency records” must include AI-generated content.
- Australia & Canada: Parliamentary committees opened inquiries into digital-record retention, inviting New Scientist editors to testify.
Ongoing Challenges: Transparency in an Era of Autonomous Drafting
Despite the victory, several challenges persist:
- Versioning: Chatbots generate multiple drafts. Which version counts as the “official” record?
- Private Devices: Ministers sometimes use personal phones for quick AI queries. Enforcing retention across devices is tricky.
- Security vs. Transparency: Releasing AI logs may inadvertently expose system prompts, which can be exploited to jailbreak models.
Policy analysts advocate a two-tiered system: public release of user-visible prompts/responses, with sensitive system instructions stored under classified annexes accessible to select oversight bodies.
Best Practices Emerging Post-Ruling
Government departments are now adopting concrete steps to comply:
- Mandatory logging of all AI interactions through enterprise accounts linked to central archives.
- Automated tagging of personal data to streamline redaction workflows.
- Quarterly audits by independent digital-governance units.
A Blueprint for AI-Age Accountability
New Scientist’s 2025 FOI crusade did more than score a scoop—it rewired transparency statutes for an AI-saturated future. By establishing that chatbot logs are public records, the magazine helped ensure that algorithmic aids do not become black boxes shielding policy-makers from scrutiny. As governments worldwide integrate AI deeper into their decision-making apparatus, the principles forged in this case—accessibility, redaction over rejection, and technological agnosticism—offer a blueprint for democratic accountability in the digital century.
