Copyright attaches to a creative work the moment it’s fixed in tangible form. The moment you write the sentence, record the note, or save the illustration file, you own the copyright. You don’t need to register, publish, or announce it.
This is the legal baseline. But where you store your work—and under what terms—shapes what other parties can do with it before you’ve made any public-facing decisions. The cloud storage service that hosts your unpublished manuscript isn’t neutral infrastructure. It’s a party with its own interests in the content you’ve uploaded.
For most creators, this distinction has never mattered. Now, in an era where the same platforms storing files are also training AI systems on user content, it does.
The License Problem in Cloud Storage Terms
When you upload a file to a cloud storage service, you don’t transfer copyright. The file is still yours. What you do grant is a license—a set of permissions that allows the service to do specific things with your content.
Every major cloud storage provider requires this license. The technical reason is legitimate: to serve your file back to you, the service needs permission to store, transmit, and display it. Without a license, even routine file retrieval would be legally uncertain.
The problem is that many licenses are written far more broadly than necessary for file delivery. Google Drive’s Terms of Service include a license to “host, reproduce, distribute, communicate, and use” your content “for the limited purpose of operating, improving, and providing our services.” Dropbox’s terms contain similar language.
Each of these companies has standard explanations for why this language is necessary and narrowly applied. But the legal scope of the license—what it technically permits—is broader than the stated intent.
For published works, this matters less: the content is already public, and your rights as the creator are established in the market. For unpublished works—a novel in its third draft, a song that hasn’t been released, a visual art series not yet shown—the gap between what the license technically permits and what you’d actually consent to is worth examining.
AI Training and Unpublished Content
The license question has become urgent because of AI training.
Large language models are trained on text. Image generation models are trained on images. Music AI systems are trained on audio. The quality of these systems depends on the volume and variety of training data. Platform operators have strong economic incentives to use their users’ uploaded content as training data, and many have updated terms to explicitly permit this.
The specific concern for creators: if your unpublished manuscript is used as training data for a language model, something of your work—its prose style, its structural patterns, its character voice—has been absorbed into a system that may generate similar content for others. Your legal copyright remains intact. Your practical originality advantage doesn’t.
This isn’t hypothetical. Adobe’s updated terms, including provisions about content use for generative AI features, triggered significant backlash from illustrators and photographers who use Creative Cloud. Platform after platform has introduced AI features powered by user content. The text is your novel-in-progress; the “improved service” may be an AI writing assistant someone else is using.
The creators most affected are those in the revision stage: work that’s substantive enough to be valuable as training data, but not yet released. Finished, published work is already public. Truly early drafts may not be substantive enough to matter. The in-between state—a completed draft, a demo recording ready for label consideration, a series of illustrations ready to be shown—is where the exposure is highest.
What Different Creative Forms Are Risking
Writers and Screenwriters
A 90,000-word novel draft stored on a major cloud platform contains extensive information about an author’s distinctive prose style: sentence rhythm, vocabulary range, dialogue patterns, structural choices. These patterns can be learned by AI systems trained on the text, even without replicating the specific story.
For screenwriters, the concern extends to original story structure, original IP, and dialogue that might be absorbed into tools used by studios and production companies. The industry is already grappling with questions about AI-generated scripts; writers with original material in cloud storage are in a related conversation.
Musicians
Unreleased recordings contain multiple categories of sensitive creative content: the composition itself (melody, harmony, rhythm), the performance, and the production choices. Each is potentially valuable to AI training for music generation, audio deepfakes, or style transfer systems.
Demo recordings shared with labels or collaborators via cloud links are particularly exposed. A link sent to a music supervisor doesn’t expire when the sync deal falls through.
Visual Artists and Illustrators
Illustrators have been among the most vocal about AI training on their work, because the relationship between training data and model output is most visible in image generation. Style—the thing that makes an illustrator’s work recognizable and commercially valuable—can be absorbed by models trained on sufficient examples.
For visual artists with unpublished series or work in progress, the question is whether storing those files on a platform that trains AI on user content exposes work that you haven’t chosen to put into the world yet.
Filmmakers
Scripts, storyboards, production design documents, and rough cut footage all represent significant creative and financial investment. Pre-production material in particular is sensitive: the script is the blueprint, and its exposure before release can affect the project’s commercial viability.
What a Privacy-Respecting Storage Policy Should Say
When evaluating cloud storage for unpublished creative work, specific policy language matters more than marketing.
On AI training: Look for an explicit statement that uploaded content is not used to train AI models that benefit third parties. The phrase “service improvement” is too broad—it often encompasses AI training. “We do not use your content to train AI” is the clearer commitment.
On the license scope: The license granted to the service should be limited to what’s necessary to deliver the service back to you: storage, retrieval, transmission to your devices. Broad language about “developing new products,” “improving services,” or “enhancing user experience” extends beyond what’s necessary for file delivery.
On deletion: If you delete your files—or close your account—what actually happens? Vague language about “reasonable time periods” for deletion leaves room for content to persist in backups, training datasets, or derived models after you’ve removed it. A clear, permanent deletion commitment is important.
On data sharing: Does the provider share your content with third parties? For what purposes? Training data is sometimes sourced from partners rather than directly—a platform might share uploaded content with an AI development partner as part of a service agreement.
Practical Guidance for Creative Storage
Separate your storage by publication status
Maintain distinct storage locations for:
- Published works: things that are already public, where the privacy calculus is different
- Active projects: work in progress that you’re actively revising
- Archive: completed drafts that haven’t been released
The most sensitive category is active projects. These deserve the highest-privacy storage.
Think about version history
Cloud storage services often maintain version histories of files—previous drafts accessible through the platform. This is useful for recovery but means that even if you delete the final draft, prior versions may persist.
For unpublished work, you might maintain your own version history locally and store only the current working draft on cloud services. This limits exposure: the version you’re actively working on is in cloud storage; earlier versions are on your local machine or encrypted local backup.
Be cautious with collaboration links
Collaboration is a legitimate creative need. But links shared with an agent, editor, label contact, or collaborator don’t expire automatically. Standard practice is to share the file, receive the feedback, and then revoke the link—but this step is often skipped.
If you use cloud storage for collaboration, build link revocation into your workflow as a deliberate step, not an afterthought.
Consider where you draft vs. where you store
Some creators write in cloud-based word processors (Google Docs, Notion) and then export to private storage. This is a reasonable hybrid: the drafting tool is convenient; the final archive is private. The limitation is that the drafting tool’s platform still sees the content during the drafting phase. For work at its most sensitive, drafting in a local application before storing is cleaner.
The Registration Question
Copyright registration doesn’t affect your ownership, but it does affect your legal remedies if your work is infringed. In the US, registration before infringement (or within three months of publication) gives you access to statutory damages—fixed amounts set by law—which are often far easier to recover than actual damages.
Registration creates a public record with the Copyright Office, separate from where you store the work privately. Some creators register key drafts periodically—particularly if the work is ready for submission and exposure to industry contacts.
Registration doesn’t fix the problem of AI training on your content: your legal copyright is intact whether or not you’ve registered. But it strengthens your position in any future dispute about unauthorized use.
What daftei Offers for Creative Work
daftei stores files of any type: documents, audio files, images, PDFs, folders of mixed formats. Files are encrypted with AES-256 at rest and TLS 1.3 in transit. daftei doesn’t sell your data, doesn’t show advertising, and never uses your content to train third-party AI models.
Storage is organized by folders, so you can create the project-and-status structure that works for your practice. Storage starts at 5 GB free, with unlimited storage on the Pro plan at $5.99/month or $44.99/year. A lifetime plan is available at $89.99.
If you close your account, daftei starts a 30-day grace window before permanently and irreversibly erasing your data. No vague “reasonable time periods”—a defined window followed by genuine deletion.
Your work belongs to you. The cloud service holding it should acknowledge that cleanly.