Does AI Steal Your Manuscript? What Authors Need to Know in 2026
Whether an AI tool trains on your manuscript depends on the specific product and plan you are using, not on AI as a technology in general. For well-designed API-backed tools, the short answer is no: your manuscript is not used for training. The nuance is that some consumer AI products operate under different defaults, and a wave of copyright litigation has made it harder to separate what the lawsuits say from what actually happens to a file you upload today.
This guide explains the distinction plainly, covers what the policies and court cases actually say as of July 2026, and tells you what to look for before uploading your manuscript anywhere.
Two distinct risks: training data and uploaded inputs
Authors worried about AI and their manuscripts are usually concerned about one of two separate things, and it matters greatly which one you are asking about.
The first risk is historical training data: books that were scraped from the internet and used to build AI language models without the authors' permission. This is what the current wave of lawsuits is about, and it concerns something that already happened, not what happens to your manuscript today.
The second risk is what happens after you upload your manuscript to an AI-powered tool right now: does the company keep your file, use it to train future models, or share it with anyone else? This is the question that matters when you are choosing a tool to use.
These two questions have different answers, from different sources, and confusing them leads to either unnecessary fear or misplaced confidence. The sections below address each in turn.
What the lawsuits actually say
The most significant settled case involving authors and AI is Bartz v. Anthropic. Brought by authors including Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, the lawsuit alleged that Anthropic used books downloaded from LibGen and other piracy sites to train its models. The settlement, worth $1.5 billion, was preliminarily approved by Judge Alsup on 25 September 2025, with a final approval hearing on 14 May 2026. According to the Authors Guild, which tracks the case closely, the settlement will be distributed among rightsholders of approximately 500,000 titles.
The critical point: Bartz v. Anthropic concerns HISTORICAL training-data scraping. The books in question were used to build the model before it was deployed, not uploaded by users of the tool afterwards. The settlement does not say anything about what Anthropic does with data you send via its API today.
The Authors Guild v. OpenAI consolidated case (Baldacci v. OpenAI, S.D.N.Y.) is ongoing as of July 2026, in active fact discovery, with bellwether trial selection expected in late 2026. It similarly concerns books allegedly used without authorisation to train OpenAI's models. No final judgment or settlement has been reached.
Both cases are important for the broader question of author rights in the age of AI. Neither tells you what happens to a file you upload to an API-backed tool today.
What the API policies actually say as of July 2026
The policies governing what AI companies do with data you send via their APIs are materially different from the policies governing their consumer products.
Anthropic states explicitly that its consumer terms updates, including any training opt-in arrangements, "do not apply to services under our Commercial Terms, including Claude for Work, which includes our Team and Enterprise plans" and explicitly exclude "Our API, Amazon Bedrock, or Google Cloud's Vertex AI." API customers are governed by separate commercial terms under which data is not used for training by default.
OpenAI's position on API data is consistent with this: data processed via the API is not used for training by default, with retention limited to 30 days for abuse monitoring unless the customer opts into an Enterprise Zero Data Retention arrangement. OpenAI's policy page was unavailable at the time of publication (July 2026), so check the current wording at openai.com/policies before relying on it; this position is corroborated by multiple secondary sources and aligns with Anthropic's own explicit API exclusion.
Both policies are time-sensitive. Treat the information above as reflecting July 2026 and check the current versions before making decisions.
Consumer versus API-backed: the distinction that matters most
If you paste your manuscript directly into Claude.ai or ChatGPT on a free or consumer plan, the terms governing what happens to that content are the consumer terms, not the API commercial terms. For consumer Claude.ai accounts on Free, Pro, or Max plans, Anthropic may use inputs and outputs to train models unless the user has opted out. This is materially different from the API position.
When you use an AI-powered tool that runs on top of one of these APIs, two layers of policy apply. The AI provider's API policy governs how the provider handles your data: and as noted above, this generally excludes training by default. The tool developer's own data policy governs what happens to your file within their system: whether they store it, for how long, and under what conditions they delete it.
This means the right question to ask is not "does Anthropic train on my data?" but "what does this specific tool do with my manuscript, and what does its underlying API provider's commercial policy say?"
Amazon KDP and AI disclosure: what you actually need to declare
Many authors worry that using any AI tool to work with their manuscript will trigger Amazon KDP's disclosure requirement. The rule is narrower than the worry suggests.
Amazon KDP requires disclosure when AI generated text appears in the published book. If AI produced words that appear verbatim in your manuscript, you must disclose. If AI was used only to assist, such as for editing, grammar, outlining, or, as with Show Me My Book, analysing and cataloguing your manuscript, disclosure is not required. The disclosure is for Amazon's internal records and is not shown on the product page.
Because Show Me My Book analyses your manuscript and returns a canon bible, with no AI-generated text entering your draft, no KDP disclosure is triggered. You remain the sole author of every word in your book.
What to look for in any AI tool's data policy
Before uploading your manuscript to any AI-powered tool, check for four things in its data policy:
Training exclusion. Does the tool explicitly state that your manuscript will not be used to train any AI model? This should be a clear, unambiguous commitment, not buried in a carve-out.
Retention and deletion. How long does the tool keep your file? Is deletion automatic once the service is delivered, or do you need to request it? Is there a deadline on that request?
Storage isolation. Is your manuscript held in private, isolated storage, or is it pooled with other users' data? Shared storage raises the risk of unintended access.
GDPR alignment. If you are based in the UK or the EU, is the tool's data handling lawful under UK GDPR or EU GDPR? Does it state a lawful basis for processing and acknowledge your right to erasure?
How Show Me My Book handles your manuscript
The commitments Show Me My Book makes on data handling are stated directly on the pricing page, where you can read the full manuscript protection section before deciding anything:
It never writes your prose, and it never trains on your book. Your manuscript is held in dedicated, private storage for you alone, never mixed with anyone else's work. It is deleted on request, and by default once your bible is delivered. The service is GDPR-aligned, with lawful basis, data minimisation, and your right to erasure. It never asks you to send your manuscript over plain email.
The Desktop tier, coming soon, goes further: your words never leave your computer. No upload required.
If you are building a story bible for a finished manuscript and want to understand how the tool works before uploading anything, the companion guide explains the full process: How to Make a Story Bible: A Practical Guide for Novelists.
Common questions
Does uploading my manuscript to an AI tool mean the company can train on it?
It depends entirely on the tool and the plan. Consumer AI products (such as Claude.ai or ChatGPT free tier) may use your inputs for training unless you opt out. API-backed tools, where the AI provider processes your data under commercial terms, are explicitly excluded from those training programmes by default as of July 2026. Always read the specific tool's data policy, not just the underlying AI company's general policy.
What is the difference between AI training data lawsuits and what happens to my uploaded manuscript?
The lawsuits, including Bartz v. Anthropic and the ongoing Authors Guild v. OpenAI, concern books allegedly scraped from the internet to build the AI models in the first place. That is a separate question from what happens when you upload a manuscript to an AI-powered tool today. API data policies govern the latter, and those policies generally exclude training by default.
Does using Show Me My Book require me to disclose AI involvement on Amazon KDP?
No. Amazon KDP requires disclosure when AI generated text appears in your book. Show Me My Book analyses your manuscript and returns a canon bible; it does not write or rewrite your prose. No AI-generated text enters your manuscript, so no disclosure is triggered.
What should I look for in an AI tool's data policy before uploading my manuscript?
Look for four things: whether the provider explicitly excludes your data from model training; how long they retain your file and under what conditions they delete it; whether your manuscript is stored in isolated private storage rather than pooled with other users' data; and whether they are GDPR-aligned if you are based in the UK or EU.
Concerned about manuscript privacy? Join the waitlist or read the protection commitments in full.
This guide is for general information only and is not legal advice. If you have specific concerns about your intellectual property rights, consult a qualified solicitor or attorney.