DesireBorn writes erotica with you, not for you. Give it a starting idea, and it writes one section at a time - you steer after each one.

Guides

How private are AI story platforms really?

The question sits, unanswered, on one of the largest lists of AI fiction tools on the internet, in the frequently asked questions at the bottom, where the answer would go if anyone had one. The lists rank tools by features and price. None says what happens to the story after you write it, because none of the tools says either.

This guide is an attempt at the answer. It has two halves. The first is the five questions that "private" actually breaks down into for a story platform, and what the industry generally does on each. It names no company, because the honest version of a comparison is one where every claim has been measured, and we have measured only our own. The second half is our answers, in terms you can check.

The five questions

How is it stored? On most platforms, a story is a row in a database in plaintext. That is the normal way to build software, and it means that anyone with access to the database, an employee, a contractor, an attacker with a stolen backup, can read every story on the platform. A minority of platforms encrypt stories at rest. Fewer still say who holds the key, which is the only part of "encrypted" that matters: a key stored next to the data is a lock with the key in it.

Who can read it? Even on a platform that encrypts, the people who run it can read your story, because the platform has to read it to write the next part. The question is what they do with that ability: whether staff read stories as routine work, whether their internal tools can display story text at all, whether there is any record of who looked. Most platforms do not say. A few claim that nobody at all can read your stories, which for a generating service cannot be true in the way it sounds, and is the claim to be most careful of.

What goes to the model company, and does it stay there? Almost no platform in this field runs its own model. Your text is sent to a company that does, and that company has its own retention and training policies. Some keep every request for a period; some train on it unless told not to; whether the platform tells them is a choice it makes and rarely mentions.

What else is recorded? Your email, your payment identity, your menu choices, your reading behaviour, and on many platforms a third-party analytics script that reports all of it to another company. In this category, which kinks a person ticked is a fact about their sex life, and an analytics vendor is a strange place for it to end up.

What does delete mean? On many platforms, deleted means hidden: the row is flagged and stays. On some it means removed after a retention period. On a few it means the key is destroyed and the text is unrecoverable. The word is the same in all three cases.

Ask those five of any platform, including this one.

Our answers

Storage. Every premise, character note, special instruction, steering line and section of prose is encrypted before it is written to our database. Each story has its own key, and the key that unlocks those keys is not in the database; it is held separately by the software that runs the engine. The practical consequence, which is the only version of the claim that can be checked: if our database were copied, stolen or leaked tomorrow, whoever held the copy would not be able to read a word of anyone's story. This is in production and has been measured on live data.

Who can read it. We can, and here is exactly when. The engine runs on our servers, so to write section nine it has to read your premise, your instructions and the eight sections so far. That happens in memory, for the duration of that one request, and the result is encrypted again before it is stored. We do not read stories as routine work, and our own admin tools cannot: no screen in this service can display the text of anyone's story. Our support tools show numbers. If we ever needed to look at content to investigate a specific report, we would say so.

We will not claim more than that. Some services describe themselves as zero-knowledge, meaning not even the operator can read your data. That is not possible for a service that writes the story for you, and we do not use the word. The encryption protects your stories where they sit. It does not, and cannot, protect them from the engine that writes them.

The model company. Writing a section means sending your text to the company that runs the model. We send those requests with data retention switched off and training on your content refused, and the request always goes to the same company; it is never rerouted to another. We do not send your email address or your account identifier with it.

What else is recorded. Your email address, which site you signed up on, when you last signed in, and counts: stories, sections, words, what the generation cost. The choices you make from our menus, the genre, the themes you tick, the explicitness level, the narrator, the length, stay in readable form, attached to your account, because that is how we learn what to build next. That is the one part of your activity that is not encrypted. Your own words are not in that category. The email address cannot be encrypted away: the sign-in system has to look it up to let you in, so who is a customer is always legible to us. What they wrote is not.

While you read, we record which country you connect from, where you arrived from, which sections you opened and how long you stayed. That goes to our own database, and there are no cookies and no advertising network on this site. On the public pages we also use Plausible Analytics, a cookieless service hosted in the EU, as a second count we did not make ourselves. It gets the page address and a few step names, never what you write, your email address or anything from inside your stories.

Delete. You delete a story or your whole account yourself, from inside the app, without asking anyone. It is immediate and cannot be undone. A deleted story takes its own encryption key with it, in the same operation; there is no copy of the key anywhere else, so the text is not merely unlisted, it is unreadable afterwards, by you and by us. What remains after an account is deleted is two rows of numbers, what your generations cost us and any star rating you gave, with your account and the story removed from both.

The sentence we put on every page compresses all of that. Your stories are encrypted at rest. We can read them to write them. Nobody else can read them.

What that sentence does not say

It does not say the model company cannot see your text while it writes; it sees it, for the request, under no retention and no training. It does not say your email is hidden from us, or that your menu choices are encrypted, or that the service is anonymous; none of those is true, and a service with a sign-in cannot be anonymous. It says what a stolen copy of our database would contain, which is the version of the promise a stranger can test.

How to check any platform, including this one

Read the privacy page, not the product page, and look for five things: the word "at rest" next to the word encrypted, and who holds the key; a sentence that admits the operator can read stories and says when; the name of the model company's retention setting, not just its existence; a list of what stays readable; and what deletion physically does. A page with all five is describing a system. A page with none is describing an intention, and in this category the difference is not academic.

Choose a platform for its five answers, or not at all. If you find one of ours to be untrue, tell us through the contact page, and do not paste your story into the form; the answer to that question is the whole point.