Recently we decided to switch to creem.io as our main payment gateway and they have an interesting mandatory feature for AI platforms like ours. We now have to run our prompts through their NSFW detection API (free for now, price will go up later). Which is really interesting because we haven't encountered another payment provider that handles censorship on their part yet.
Every image and video prompt goes through it before generation runs. If it flags something, the job stops there. The user sees a block on our side, but the check itself is coming from the payment stack, not from the model provider.
We've used other gateways before. They cared about chargebacks, fraud, that kind of thing. Normal payment stuff. None of them asked us to screen prompt text on their behalf. Creem is a merchant of record, so they're taking on more liability than a regular processor would. From their angle it makes sense. Fewer bad prompts, fewer disputes, fewer problems downstream. Still, it's a weird place for a content filter to live.
We always knew that AI was going to get more and more censored. Models refuse things. Platforms add policies. Studios push back on likeness and IP. That part wasn't surprising. What we didn't expect was this move from an MoR. The censorship layer used to sit at the model or at the platform. Now there's a third spot in the chain, and you don't really control it if you want to keep taking payments through them.
It's not optional either. If you run an AI image or video product on Creem, you implement their moderation API. We did, because we need the gateway. But it does make you think about where this goes if more MoRs start doing the same thing.
Are we moving towards heavier censorship in the AI industry? Feels like it. Not always in the loud way people argue about online. Sometimes it's quiet infrastructure. A check that runs before the GPU, billed per call, tied to your ability to charge a card at all.