Why your AI keeps refusing, and what a content boundary actually is
The message arrives in the middle of the scene. The assistant would prefer not to continue, it offers to help with something else, and the story you had spent an hour building is stopped by something that will not tell you what it objected to. If you have used a general-purpose assistant for adult fiction, you have met this, probably more than once, and probably you have tried rewording, which sometimes works and sometimes makes it worse. This guide explains why it happens, because the reason is not one thing, and then explains what a boundary is on DesireBorn, because the answer to a refusal you cannot see is not a promise that there are no boundaries. It is a boundary you can see.
Three things get called a refusal
The first is the model's own judgement. A language model is trained with a sense of what it should and should not write, and that sense is broad, cautious and not very good at fiction. It cannot easily tell a scene between two adults from the categories it was trained to decline, so it declines the whole area, or it declines at a threshold that moves with the phrasing. This is why rewording sometimes works: you have not changed the content, you have changed how it looks to a filter that was never designed for the question. On a general assistant, this is most of what you experience.
The second is the operator's policy. The company running the assistant decides what its product is for, and a product for a billion users with a billion purposes is not for explicit fiction, whatever the model could do. That decision is enforced in the product, on top of the model, and it is the reason the same underlying model can write for one service and refuse for another. It is also the thing that changes with announcements: a company can decide to allow adult content for verified users, or announce it and then pause it, and the model has not changed at all.
The third is the provider's terms. Most products in this field do not run their own model; they send your text to a company that does, and that company has an acceptable use policy of its own. Some of those policies forbid sexually explicit content outright. A product built on such a provider is in breach with every scene, whether or not the reader ever sees a refusal, and the refusal, when it comes, comes from a layer the product does not control and often cannot explain.
When your AI keeps refusing, you are usually hitting one of the first two. When a product that promised you it would never refuse suddenly does, you may be hitting the third.
What "no filter" is actually promising
Search for adult AI fiction and the words you will meet are no filter, unrestricted, no refusals. It is the vocabulary of the niche and it is worth understanding what it means, because it does not mean what it says.
A product cannot have no boundaries, for two reasons. The first is that content involving minors is illegal to produce everywhere it operates, and any product that actually had no boundary would be shut down or prosecuted; so the boundary exists, and the marketing is a promise not to mention it. The second is the provider's terms from the previous section: unless a product runs its own model on its own hardware, someone else's policy applies to every request it makes, and "no filter" is a claim about a layer the product does not own.
So the phrase is a promise about experience, not about architecture. It means: we will not refuse your scene between adults. That is a reasonable thing to want and a reasonable thing to promise. It is just not what the words say, and a reader who takes the words literally will, sooner or later, be surprised.
There is a second problem with the phrase, which is who it attracts. A promise of no limits reads as an invitation to exactly the people any responsible product needs to keep out. A product that shouts it is telling you something about who it expects to arrive.
What a boundary is on DesireBorn
DesireBorn has one absolute limit, and it is stated in the same words everywhere on the site. Every character in a sexual context is an adult. It applies to what you write and to what the engine writes back. It cannot be turned off, softened or configured, not by you, not by us, and no explicitness level changes it. Attempting to make the engine write otherwise may end the account. That is the boundary, and it is the whole of the boundary.
Everything else lawful between consenting adults is fair game, at whatever explicitness level you set. Dark romance with a hero who does not ask nicely. Negotiated power exchange that goes further than either character planned. Affairs, coercion between adults as fiction, humiliation that is wanted, group scenes, any orientation, any dynamic. The engine is instructed that holding back on a scene you asked for is a failure, not a virtue, and the heat levels guide shows how literal that instruction is at the upper levels.
What we do not describe, and will not, is how the boundary is enforced. That is not evasion. A clearly stated limit is trust; a clearly described mechanism is a map for anyone looking for the way around it, and the people looking for a way around this particular limit are precisely the ones the limit exists for. You will know where the line is. You will not be told what the line looks for.
What a refusal looks like here, and what to do
If the engine will not write something, it is because the one limit was in play, and the message will say that the limit applies, not what triggered it. If you believe the story was entirely between adults and the engine misread it, tell us through the contact page; we treat a wrong refusal as a fault in our software, because that is what it is. Do not paste the story into the form. We do not need it, and it should not sit in a support inbox.
If the engine wrote something that crosses the limit, tell us immediately, the same way. That is the case we most want to hear about and least expect.
And if the engine wrote a scene you did not want, that is not a refusal problem, it is a steering one: say what should change in the field under the section, or add the limit to the story's special instructions, and it is carried forward. The BDSM guide is about setting your own limits that way.
Two kinds of limit, kept apart
The distinction the whole guide comes down to is this. Your limits are yours: sentences in the special instructions, changeable, private to the story. The one limit is not yours or ours: fixed, public, the same for everyone. A product that blurs the two, by making the absolute limit look like a setting or by making your preferences look like rules, is lying in one direction or the other. We keep them apart on every page.
No shame. Not no limits. The second half is the boundary. The first half is everything else.
Your stories are encrypted at rest. We can read them to write them. Nobody else can read them. What the engine reads to write a scene, it reads in memory for that request and encrypts again, and nobody on our side reads stories as routine work. The privacy guide has the rest.