The Government File That Became A Prompt - Affan Yesvi

It is 5:30 p.m. A government office. One file remains on the desk.

A policy draft needs polishing. A note must be shortened. A presentation awaits the next morning. The officer opens a generative AI tool. Types:

"Please make this document concise and improve the language."

The document is pasted. Within seconds, the job is done.

Nothing seems to have gone wrong. No hacker. No deliberate leak. No malicious intent. Just an employee trying to finish the day's work.

But there is one question worth asking: Where did that document go?

This may sound like a minor concern amid the AI excitement. It is not.

AI has become an extraordinarily useful administrative tool. It can draft correspondence, summarise reports, translate material, analyse information and reduce hours of repetitive work to minutes. For government, where paperwork is often synonymous with administration, the attraction is obvious.

But the same convenience creates a new and poorly understood confidentiality risk.

A File Can Now Leave Through a Prompt

For decades, governments built elaborate systems to protect official information. Files were marked confidential. Access was restricted. Sensitive documents circulated through defined channels.

Digitalisation changed how information moved. AI changes it again.

A document no longer needs to be emailed, photocopied or physically removed from an office. It can leave through a prompt. And the person entering it may not even consider that an act of disclosure.

This is not to suggest that every AI platform makes everything entered into it public. That would be inaccurate and alarmist. Different services come with different privacy arrangements, retention policies, security controls and enterprise protections.

But that is precisely the point. Does the government employee know which system they are using, how it handles data and whether the department has authorised its use for that information?

That question should become as routine as asking whether a document is confidential.

If Corporations Are Cautious, Why Isn't the Government?

Here lies an irony. While government offices increasingly use AI as a convenient administrative assistant, many large corporations have restricted the use of public or unauthorised AI tools for confidential work.

They have not stopped using AI. Rather, they are moving towards controlled, enterprise environments where sensitive information can be better protected. The reason is simple: their data is their competitive advantage.

A product roadmap, marketing strategy, financial projection or technical design can be worth millions. Companies therefore worry about where their information goes, who can access it and how it may be processed.

The lesson for government is obvious. If corporates are cautious about protecting their intellectual property, should the government be any less careful with policy documents, procurement strategies, citizen data and information concerning national interest?

The question is not whether AI should be used. It is where, how and with what safeguards.

Convenience cannot come at the cost of confidentiality.

Consider a Government Tender

The risk becomes clearer in procurement. Before publication, a government tender may contain technical specifications, eligibility criteria, evaluation parameters, estimated costs and the department's thinking about how competition should be structured.

An officer may paste the draft into an AI system and ask: "Improve these tender conditions." Another may ask, "Are these technical specifications too restrictive?" Someone else might ask, "Suggest better eligibility criteria."

The AI may provide a brilliant answer. But should an unpublished tender have been entered into that system in the first place?

The same question applies to draft policies, internal financial proposals, legal opinions, infrastructure plans, examination papers and strategic notes. Something does not have to be classified "Secret" to deserve protection. Confidentiality exists on a spectrum.

India Cannot Afford to Ignore This

India is rightly embracing artificial intelligence. The Government of India has invested heavily in building an AI ecosystem through the IndiaAI Mission, approved in 2024 with an outlay of ₹10,371 crore over five years. By July 2026, the government reported 762 AI use cases identified across 62 ministries and 543 Data & AI Labs.

More than 1.7 crore government employees had been onboarded on iGOT Karmayogi by July 2026, with hundreds of courses covering AI and emerging technologies. In April 2026, the Ministry of Electronics and Information Technology set up the AI Governance and Economic Group as a high-level inter-ministerial mechanism for national AI governance.

All of this is welcome. But it raises another question: Are we teaching government employees how to use AI faster than we are teaching them how to use it safely?

Other Governments Have Already Acted

This is not a concern unique to India. The UK Government's guidance to civil servants explicitly cautions against putting sensitive or personal information into publicly available generative AI tools. It warns that doing so could compromise information that is not yet public, including information revealing government intentions.

Canada has taken a similarly clear approach. Its guidance tells public servants not to enter protected, classified or personal information into public generative AI tools and warns about risks arising from how suppliers may handle or retain input data.

The lesson from these countries is not that governments should stop using AI. It is that AI adoption must come with information-security rules.

Don't Ban AI. Govern It.

There is little point in telling government employees simply, "Don't use AI." They will use it. And they should, where it can make government more efficient.

The answer is to establish clear departmental rules around the use of AI for official work, including which platforms are approved, what categories of information can be entered, what safeguards are required and what information must remain outside external AI systems.

Most importantly, every government employee using AI for official work should receive basic AI confidentiality training. Not a technical lecture. A practical one.

Where does the data go? Who controls it? How long can it be retained? What is the difference between processing and training? What information can be safely anonymised? What should never be entered? Which platforms are approved?

One simple principle could become part of government culture:

If you would not send a confidential document to an unknown third party, don't paste it into an unapproved AI chatbot.

The Real Issue Is Trust

Ultimately, this is not just a technology debate. It is about trust.

Citizens entrust government with information. Businesses submit confidential commercial material. Departments develop policies and strategies long before they become public. Government holds information whose premature disclosure can have consequences far beyond an individual office.

AI can help the government handle this information better. But it cannot become an excuse to handle it casually.

The objective, therefore, should be simple: How do we use AI without compromising the information entrusted to government?

The AI revolution is already here. Government will—and should—use it. But AI literacy must now include confidentiality literacy.

Because the next breach of government information may not start with a hacker stealing a file. It may begin with an ordinary employee, an ordinary document, an ordinary mobile phone and an innocent request:

"Can you improve this?"

And then, with one press of Enter, the government file becomes a prompt.

About the Author: Affan Yesvi is a writer, thinker, and motivational speaker passionate about exploring life's many dimensions. His work focuses on personal growth, human experiences, and driving meaningful social impact through heartfelt advocacy.

Disclaimer: The views and opinions expressed in this article are those of the author(s) and do not necessarily reflect the official policy or position of the publisher. While every effort has been made to ensure the accuracy of the information, the publisher is not responsible for any errors or omissions, or for the results obtained from the use of this information.