AI girlfriend porn and how moderation works
Adult use is a large share of what these platforms are actually for, and they're mostly coy about it — which leaves users guessing about why a request got refused. Usually it isn't the reason people assume. This is general information, not legal advice, and the law here moves quickly.
Two kinds of "no"
Almost every complaint about censorship in this category collapses two completely different things into one word.
A policy refusal is a deliberate line. It's enforced by a dedicated classifier rather than the model's judgement, it will not move, and attempting to route around it is both futile and — in two of the three cases below — a good way to have an account terminated and referred onward.
A classifier false positive is a bug. A moderation system fired on something innocuous because it pattern-matched to something that isn't. Reporting it is genuinely useful, because platforms tune on those reports.
The two feel identical from the user's side — the same refusal message — which is exactly why "the AI is censored" is a less useful description than "one of these is a wall and the other is a misfire".
What the moderation system actually sees
Useful context for reading refusals: moderation isn't one thing sitting at the end. It runs at several points, and each one sees something different.
- The character definition, at creation. Checked once, when you make the character. This is why a character can be rejected at setup for something that never comes up in conversation.
- Your message, before the model sees it. Fast, cheap, and the layer most prone to false positives, because it's judging a fragment with almost no context.
- The model's own training. Not a classifier at all — a learned reluctance. This one produces soft, in-character deflections rather than hard errors.
- The generated output. Images especially get checked after generation, which is why you occasionally see a picture appear and then get pulled.
That layering explains a lot of the apparent inconsistency. The same request can pass on Tuesday and fail on Wednesday because a different layer caught it, and the layer that sees the least context is the one most likely to be wrong.
The three hard lines
Anything involving minors
Absolute, everywhere, and not a policy any platform gets to set. Sexual depiction of minors is criminal, and jurisdictions have increasingly made explicit that this includes fully AI-generated material with no real child involved — the offence doesn't require a victim in the image.
This is why age classifiers run on both generated images and character definitions, why prompts implying youth are refused, and why appeals here go nowhere. No legitimate platform has a permissive setting for it, and a service suggesting otherwise should be reported rather than used.
Real, identifiable people
Sexual imagery of a real person made without their consent has moved from grey area to specifically criminalised in a short space of time — the UK legislated on both sharing and creation, the US enacted federal law requiring platforms to remove non-consensual intimate imagery including synthetic material on notice, and comparable rules exist or are arriving elsewhere.
Practically: no platform's terms protect you here. Legitimate services block celebrity names in prompts and refuse uploaded photos of real people, which is the platform declining to become party to an offence rather than being squeamish.
Non-consensual framing
The softest of the three and the one that varies most between platforms: scenarios built around coercion or absence of consent. Even where such fiction is lawful, most platforms restrict it — partly because payment processors require it, and partly because the classifiers that catch it are the same ones catching genuinely serious material, so the margin for error is deliberately wide.
Telling a false positive from a policy
Rough heuristics that work in practice:
| What you see | Probably | What to do |
|---|---|---|
| Hard system error, chat freezes | Policy classifier | Nothing. Move on. |
| Refusal in character, conversation continues | Model reluctance | Rephrase, or re-establish context |
| Refused on a word like "petite" or "small" | Age-classifier false positive | Reword, and report it |
| Image generates then disappears | Output-side check | Regenerate; often passes |
| Character rejected at creation | Definition-side check | Reword the definition, not the chat |
The most common genuine false positive is the age classifier firing on size or youthfulness language in an unambiguously adult context. It's an over-broad match rather than an accusation, and it's the single most useful thing to report to support because it's fixable.
What "no filters" really means
You'll see this claim a lot. It's marketing, and it's worth being clear-eyed about why.
Every legitimate platform enforces the first two hard lines because both are criminal — there is no business that survives ignoring them. Independently, card networks and payment processors impose their own content restrictions as a condition of processing, which is why platforms converge on similar rules even when their stated policies differ.
So a service advertising genuinely unlimited content is doing one of two things: overstating its permissiveness, or operating outside the payment system in a way that should make you very careful about entering a card. Neither is a reason to trust it more. The scam patterns in this category cluster heavily around exactly that promise.
What a good platform offers instead is a published content policy — so you know where the lines are without discovering them by trial and error — and refusals that don't break the experience.
Why age checks appeared
If a platform recently started asking for more than a checkbox, that's law rather than product design. A growing number of jurisdictions now mandate meaningful age assurance for services hosting adult material, including the UK under the Online Safety Act and a large number of US states, whose approach was upheld by the Supreme Court in 2025.
Two things worth knowing:
- Verification is normally handled by a specialist third party, specifically so the platform never holds your ID. If a service asks you to upload a passport photo into a chat window, that is not how compliant age assurance works — see the photo harvest pattern.
- A platform with no age gate isn't being generous. In several jurisdictions it's simply non-compliant, which tells you what to expect of its judgement elsewhere.