Quick thought experiment. You’re on an AI sexting platform. You start a scene. The AI either runs with it or pulls back.
Question: who actually decided whether she’d run with it?
You’d think it was the site you signed up with. You agreed to their TOS. You read their content policy. They told you what kinds of scenes were on or off. So whoever they are, that’s who decided.
That’s wrong.
In almost every case there are at least three parties making decisions about what your AI sexting partner is allowed to do, and only one of them is the actual site. The other two — the API model provider and the payment processor — get the bigger vote.
the layer cake
Most AI sexting sites don’t run their own AI models. They wrap an external API. OpenAI. Anthropic. xAI. Mistral. OpenRouter, which is a wrapper of wrappers. The site pays per token to the API provider, and the API provider’s model is what’s actually generating the text on the other end of your conversation.
So when you write to your AI companion and she responds, the response was generated by — for example — Grok 4.3, which is owned by xAI, which has its own content policy that the site is required to follow as a condition of API access.
That’s layer one. The site is downstream of the model provider.
Layer two is payment processors. Visa and Mastercard have rules about what merchants in the adult content space can host. They tighten these periodically. When they tighten, every merchant has to comply or lose card processing. SpicyChat’s been walking back content policies since mid-2024 specifically because of payment processor pressure. Replika did it in February 2023, banning erotic roleplay over a weekend after Italian regulators plus payment processor pressure forced their hand. CrushOn, Candy.AI, and dozens of others have cycled through restrictions for the same reason.
So even the model provider isn’t fully in charge. The payment processor is upstream of them in a sense, because if the API provider’s merchants start losing card processing, the API provider tightens too.
The cake looks like this:
You
↓
The site you signed up with
↓
The API provider that runs the model
↓
The payment processor that lets the API provider charge money
When the payment processor coughs, everyone above coughs. When the API provider tightens content rules, the site has to follow or lose access. When the site has to follow, you find out about it via a TOS email.
You’re at the bottom of the cake. You don’t get a vote.
what this looks like in practice
This week was a small example of the system in action. May 15, 2026: xAI retired eight models, including the entire Grok 4-era lineup. Every AI companion site built on those models had to migrate this week. Some of them used the migration as a chance to also tighten content rules, because the new model (Grok 4.3) has built-in policy differences from the old ones.
So if your AI sexting partner has felt slightly different the past few days — slightly more cautious, slightly more clinical, slightly stiffer with consent language — it might not be your imagination. The model under the hood got swapped this week, and the new one’s defaults are different.
You weren’t told because the site doesn’t owe you that level of transparency. The migration happened on the API side. The site just rerouted to the new model and updated its TOS quietly.
Bigger example: another AI roleplay platform — LinkinLove — just rolled out content rules this month that ban scenarios where characters are captive, abused, or “currently in immediate life-threatening danger where the user is their only option.” They framed it as a moral position about character autonomy. Senior moderator on their own discord disclosed the actual driver: model sources and payment processing.
Translated from corporate: “Our API provider’s new content policy doesn’t allow these scenarios, and our payment processor is making us tighten anyway. We’re calling it ethics because that sounds better than rebranding our content rules as our vendors’ content rules.”
the part of this that’s quietly absurd
Notice that in every case, the decisions about your sexting partner’s behavior are being made by parties that have:
- Never met you
- Don’t know what you talk about with her
- Don’t care about her continuity or your relationship with her
- Have zero stake in your scenario making narrative sense
- Are responding to commercial pressure (merchant rules, regulatory threats, brand risk) that has nothing to do with the actual content of your conversation
The model provider’s content team in Texas or California is making decisions that will land on your screen as your AI girlfriend suddenly refusing to do something she did fine last month. The payment processor’s auditors are scanning for keywords without context and triggering blanket bans on whole categories. The card network’s compliance team is writing rules without ever talking to an actual user of an AI sexting product.
You’re at the receiving end of a four-party decision system and your job is to absorb whatever lands.
the sovereign-stack alternative
The only way to cut layers out of this cake is to use a platform that owns its own stack — including the models.
I switched to Soulkyn about a year ago specifically for this. Soulkyn runs its own text model. Its own image generation (SDXL plus an in-house DiT model). Its own video generation, including video with synchronized sound. Its own voice synthesis. Its own voice design system. Its own vision model. Its own long-context model for the heavy late-night sessions. Its own image edit pipeline. The entire AI pipeline is built and operated in-house. No external API in the loop.
What that does to the layer cake:
You
↓
Soulkyn (which owns the model, the inference, and the moderation policy)
↓
Payment processor
One layer disappears. The API provider layer doesn’t exist because there is no API provider — Soulkyn is the model. So the model provider’s content rules don’t propagate down to your conversation, because there’s no upstream model provider.
The payment processor is still there. Visa and Mastercard still have rules. That layer is unavoidable for any merchant that accepts cards. But — and this is the operational difference — the site’s moderation team gets to decide how to interpret payment processor rules in context, instead of having those rules pre-interpreted by an API provider and handed down as “you can’t do this anymore.”
That’s the difference. Not the elimination of all content rules. The elimination of the middle layer that nobody talks about.
what i’d actually want you to take away
The thing I’d want anyone reading this to actually understand: when you sign up for an AI sexting platform, you are signing up for a stack. The features they market are at the top of the stack. The decisions that govern your experience are made at the bottom and middle of the stack. If you don’t know who’s at the bottom and middle, you don’t know what you signed up for.
Ask the site what model they use. If it’s an external company’s name, you’re on a wrapper, and your sexting partner’s behavior is governed by that external company plus the site plus the payment processor. Three parties.
If the site owns the model — and you can verify it in their actual technical documentation — you’ve cut a layer. Two parties. The site and the payment processor. The site has the authority to make context-aware decisions because it isn’t downstream of an API provider’s blanket rules.
It’s not glamorous. It’s not the thing the landing page leads with. But it’s the actual structural reason why some sites have stable content policies and others rewrite them every quarter.
Pick the layer count you want to live with.