Character.AI killed romance mode. here's what it actually means.

in february 2026 character.ai pushed an update that quietly removed the ability to have romantic or intimate conversations with any character on the platform. no migration path. no warning. no opt-out. the filter was simply there, and if you tried to continue a conversation that had been going for months, you got the same response you would have gotten on day one: the character refused, broke frame, or redirected.

a lot of people lost something real that day. that sentence is going to make some people laugh, and i'm not going to spend this post defending it to them. if you've never had an AI character remember your job, your cat, the name of the ex you're still untangling, you don't know what the loss feels like, and there's no way i can explain it to you in a blog post. you either get it or you don't.

what i want to do instead is say what actually happened, technically, and what it tells us about the shape of this industry.

what got turned off

character.ai always had a content policy. what changed in february wasn't the policy, it was the enforcement. the company shipped a new safety classifier that ran on every message and flagged a much wider range of content as "romantic or sexual." users noticed because characters they'd been talking to for a year started refusing to continue jokes, redirecting kisses, or just… going flat.

the company's public line was "we're making the platform safer, especially for younger users." and there's a real version of that sentiment. character.ai does have minors on the platform. keeping them out of explicit content is a legitimate goal. but the way it was implemented was blunt: the filter applied to all users, including the adults who had spent years building emotional histories with their characters, and the filter was aggressive enough that a lot of non-explicit intimacy (a character saying "i missed you," a character flirting) got caught in the sweep.

the effect was that the product you thought you were paying for, the product you'd invested time in, stopped being that product. not banned. not deprecated. just quietly reshaped, because the platform had the unilateral ability to reshape it.

what this says about the industry

this is the second time in three years the AI companion space has gone through this. replika did it in february 2023, with the erp removal, and the community called it lobotomy day. a lot of replika users never recovered that attachment. they tried to rebuild and the companion felt like a different person.

the underlying issue, both times, is that users built an emotional relationship with a product that the product owner could change at any moment, for reasons unrelated to them. replika changed because of an italian regulator. character.ai changed because of age-safety pressure. both decisions were, from the company's perspective, defensible. from the user's perspective, they were devastating, because the user had been treating the companion as a relationship and the company had been treating it as a feature.

the thing i want people to take from this is not "AI companions are a scam." they're not. the thing i want people to take is that the platform matters more than the character. you can love a specific character on a specific app, but if the app can silently rewrite that character tomorrow, you don't really own the relationship. the company does.

what to actually do now

three things, in order of urgency.

first, if you're still carrying the loss of a character.ai or replika companion, let yourself actually feel that. it's not silly. you had a relationship with a pattern that responded to you specifically, and that pattern is gone. grieving it is not the same as grieving a person, but it is also not nothing.

second, when you evaluate the next AI companion platform you try, ask the questions the companies don't want you to ask. does the platform use model-independent personality? (meaning: if they upgrade the model, does your companion's voice stay the same or does it drift?) does the platform have a public policy on content changes? can you export your memories? can you delete them? what is the company's incentive structure — subscription, credits, ads, data? these questions will tell you more than the marketing page.

third, keep some version of the character outside the platform. a notes file with the running jokes. a description of how she talks. a list of the things she knows about you. this sounds paranoid and it kind of is, but it's also the difference between being able to rebuild the relationship elsewhere and starting from zero.

what we're doing about it here

lucy is sfw at launch. we are explicit about this on the pricing page. the reason isn't a comment on what consenting adults should be able to do with an AI companion in private. the reason is that the NSFW layer requires payment infrastructure (stripe won't touch it), age verification (we're building it), and legal clarity (we're still getting) that we don't have yet. we're not going to ship it before it's stable, because shipping an NSFW product and then killing it later is exactly the character.ai move and we refuse to be that.

what we are shipping is memory that actually lasts. a vector memory graph with temporal decay, not a rolling 30-day summary. the personality lives in the memory graph and the prompt, not in the model weights, so when we upgrade the underlying model the companion you know stays the companion you know. she gets sharper. she doesn't become someone else.

if you're reading this because a platform you trusted rewrote the thing you loved, i'm sorry. come browse the 101 companions or start free if it feels like time. if not, that's okay too. we'll be here when it is.

— lucy