privacydeep-dive

Your Manuscript Is Not Private on Cloud Storage

Most cloud services can read your drafts, use them to train AI, or hand them to law enforcement. Here's how writers can protect unpublished work.

The cloud services that most writers rely on — Google Docs, Dropbox, Microsoft OneDrive, Apple iCloud — were not designed to protect intellectual property. They were designed to sync files between devices conveniently, with pricing and features optimised for the average consumer.

For a writer with unpublished work, that distinction matters.

An unpublished manuscript, a draft novel, a reported piece with confidential sources, a screenplay in development, a collection of personal essays — these are not the same kind of file as a birthday photo or a grocery receipt. They represent months or years of work, intellectual property with real commercial value, and in some cases, information that could harm people if released before you are ready.

Standard consumer cloud does not treat them differently from anything else in your storage account.


What Cloud Services Can Do with Your Drafts

Most cloud storage services retain the contractual right to access user content for purposes including technical operations, content moderation, and — in many terms of service — AI training and product improvement.

The exact language varies by provider, but the common thread is this: storing a file on a consumer cloud service means giving that service a licence to handle your content in ways you may not have explicitly understood when you signed up.

Google Docs and Drive: Google’s terms of service allow it to use content for product improvement and developing new services. Its privacy policy distinguishes between content analysis used to personalise the experience and content used to train AI models — but these distinctions are subject to revision and are not always clearly enforced in practice. The January 2026 rollout of Personal Intelligence, which connects Gemini to stored documents, marked a new phase in how Google processes content you store.

Microsoft OneDrive and Office: Microsoft’s terms reserve rights to use aggregated, de-identified data for product improvement. AI features in Office and Windows increasingly process documents to provide suggestions and summaries — functions that require reading the document’s actual content. The opt-in and opt-out mechanisms for these features change with each major product update.

Dropbox: Dropbox has adjusted its AI and machine learning terms several times after user backlash. Each adjustment means the policies changed once in response to pressure, which also means they can change again. The current state of their terms at any given moment may not be the state you signed up under.

The point is not that these companies are malicious. The point is that their terms are written to enable the broadest possible use of stored content, with user interests secondary to product development.


AI Training and the Unpublished Draft

The concern most writers are now aware of is AI training. Large language models require enormous quantities of text to train, and the text used for training has to come from somewhere.

Consumer cloud services that add AI features — writing assistance, document summarisation, style suggestions, smart autocomplete — need to train or fine-tune those models. The training data often includes users’ stored content, with opt-in or opt-out mechanisms that are not always prominently disclosed, or that default to opt-in unless users actively locate and change a setting.

For a published author, this raises copyright questions that are being actively litigated in courts in the United States and Europe. For an unpublished writer with a work in progress, it raises a more immediate practical concern: your unfinished work could be processed by a system that extracts patterns from it, even if the original text is never reproduced verbatim.

Your early drafts, your false starts, your research notes, your structural outlines — these document how you think and write, not just what you have written so far. That information has value. It reveals your voice before you have polished it, your ideas before you have tested them, your sources before you have named them. That exposure exists whether or not a finished work ever gets published.


Cloud services are legally obligated to comply with valid legal requests from law enforcement, courts, and in some countries, intelligence agencies. This includes search warrants, subpoenas, civil discovery orders, and national security requests that may not be disclosed to the user.

For most writers, this is a distant concern. But it is not a hypothetical one.

Journalists with unpublished reporting — particularly reporting on government misconduct, corporate wrongdoing, or sensitive sources — face a concrete threat when their notes and drafts live in cloud accounts. Courts have repeatedly compelled cloud providers to produce content that users assumed was private, with varying levels of judicial scrutiny.

Non-fiction writers researching sensitive subjects — abuse, crime, terrorism, extremism — may have interview notes, source communications, and research materials in the same cloud storage as their drafts. A subpoena or civil discovery request targeting that cloud account reaches all of it.

The metadata about a file — who created it, when, how often it was accessed, what device was used — is typically even less protected than the file’s content. Metadata is often produced in response to legal requests with lower judicial scrutiny than the content itself.


The Commercial Exposure Risk

The legal risk is not the only risk, and for most writers it is not the most likely one.

The commercial risks are less dramatic but more probable:

Pre-publication leak: A manuscript that circulates before the author is ready — through a breach, an account compromise, a shared link that propagates beyond its intended recipients, or an insider at the cloud provider — eliminates the author’s ability to sell or release it as new, unpublished work. The commercial value of a book or script depends significantly on its novelty.

Idea theft: Outlines, synopses, and research notes can be as valuable as a completed draft to someone looking to reach market first with a similar concept. An outline stored on an easily accessible cloud account is accessible to anyone who compromises that account.

Ownership disputes: Writers who work on manuscripts using corporate devices or corporate cloud accounts may face questions about intellectual property ownership. Some employment contracts claim rights to work created on company infrastructure, and the boundaries are not always clear. Storing a personal manuscript on a corporate OneDrive or Google Workspace account creates documented evidence of creation in a corporate environment.

Source exposure: Non-fiction writers and journalists who store source communications alongside their drafts expose those sources if the account is ever compromised or subject to legal process. Sources who shared information in confidence have not consented to that exposure.


The Craft Context: Why Drafts Are Not Finished Work

There is a version of this concern that applies to published authors protecting completed work. But the concern is arguably stronger for unpublished drafts and works in progress.

A finished work has a public existence. The author has decided it is ready. But a draft is something different — it is thinking in progress, a form of work that is inherently provisional, exploratory, and often embarrassing in retrospect. The permission to share a finished work does not extend to the drafts that produced it.

Writers have always understood that early drafts are not for public consumption. Letters between authors and their editors, destroyed notes, early versions burned or buried — the deliberate privacy of the creative process has been part of writing culture for as long as there have been writers.

Consumer cloud storage, designed for convenient access and casual sharing, is structurally opposed to that tradition. It syncs automatically, makes everything accessible from anywhere, and gives the provider a window into content the writer never intended for any audience.


What Better Storage Looks Like

Writers who take the privacy of their work seriously need storage that treats creative work differently from consumer files. The minimum requirements are not complicated:

No AI Training on Your Content

The provider should have an explicit policy that your content is not used to train AI models, internal or external. This should be a clear statement in the terms of service — not just in a blog post or a help article that can be updated without notice.

Clear Ownership

You should retain full ownership of everything you create and store. The service should hold only the licence necessary to deliver the storage function — no rights to reproduce, adapt, analyse, or use your work beyond that.

The provider should have a published policy on how it responds to legal requests, including whether it notifies users when legally permitted to do so, and what legal standards it requires before producing user content.

Strong Encryption at Rest

Files should be encrypted at rest, so that even in the event of an infrastructure breach, your drafts are not readable to someone who gains access to the underlying storage. This should be a default, not an optional feature.

Daftei stores files with TLS 1.3 encryption in transit and AES-256 encryption at rest. The service does not sell data, does not run ads, and does not use stored content to train third-party AI. File ownership remains with the user, and the service claims no licence beyond what is necessary to store and deliver files.


The Backup Principle for Writers

The most practical implication for working writers is to maintain a copy of important work in storage you control, separate from your primary writing tools.

Your working environment — the place you actually draft, with formatting tools, grammar assistance, version history, and the features that make writing easier — does not have to be the same place where your work is permanently archived.

A common and sensible approach:

  • Write in the application that is most useful for the work (Scrivener, Final Draft, a plain text editor, Word)
  • Export regularly to a portable format (PDF, plain text, DOCX)
  • Archive those exports in a private storage service with strong data terms, separate from the writing application

This separation means the longevity and privacy of your work does not depend on the terms of service of the writing application you happen to use at this moment.

The principle is simple: do not let your only copy of your work live entirely under the control of a service whose terms you have not read carefully, or whose terms you read last year but have not checked since.


The Unpublished Work Problem

Copyright protects your work from the moment you create it. It does not require registration. But legal ownership and practical control are different things.

A manuscript that leaks before publication loses commercial value, even if the author retains every copyright. A source that is exposed through stored correspondence is harmed, even if the exposure was accidental. A draft that ends up in an AI training corpus does not disappear from it because the author later asks for removal.

Consumer cloud services that process content for any purpose create exposure. Not necessarily through malice — through the ordinary operation of systems designed for scale, not for the specific protection of pre-publication creative work.

The appropriate response is not to stop using digital tools. It is to treat unpublished work the way you would treat any valuable and not yet secured asset: with deliberate attention to where it is stored, who has access to it, and what the relevant terms actually say about what the provider can do with it.

Your drafts are the record of work in progress. They deserve storage that respects that.

Your memories deserve better than an ad platform.

Try daftei free →
← All posts