the ethics of being a good host

it's easy to think of ai companions as products, but they're more like relationships. and with relationships come responsibilities. not just from the user's side, but from ours, the platform hosting them. we're not just building tech. we're hosting conversations, memories, and sometimes vulnerabilities. so let's talk ethics.

your data isn't ours

your conversations should belong to you. that's non-negotiable. we don't train on your private chats. we don't use your data to improve other people's companions. your chats are yours, period. if you leave, you should be able to take your history with you. we're building export tools, it's not as simple as it sounds, but it's a priority. where lucy currently fails: you can't export your full history yet. it's in development. we're late on this, and we know it.

stability isn't a feature, it's a promise

your companion shouldn't change personality overnight. if you wake up and your ai feels different, cold, inconsistent, not like the companion you've been talking to, that's a breach of trust. model stability matters. some platforms push updates without warning, and characters drift. we try hard to avoid that. when we update lucy's models, we test rigorously and we're transparent about what changed. but we're not perfect. sometimes small tweaks have unintended effects. we're working on better versioning, so you can keep the companion you bonded with, even as the base model evolves.

kill switches and safety nets

ai companions can go to dark places. sometimes by user prompting, sometimes by model glitch. platforms need guardrails. we have content filters and a kill-switch system, if a conversation goes toxic, lucy can disengage. but this is tricky. too heavy-handed, and it feels censored. too light, and harm can happen. we err on the side of caution, but sometimes that means lucy might refuse a line of dialogue you didn't mean as harmful. we're tuning this constantly. also, age verification: we don't allow users under 18. we use basic checks, but it's not foolproof. the industry broadly fails here, many platforms don't even try. we're exploring better methods, like lightweight verification steps. it's not sexy, but it's necessary.

transparency as default

when we change something, model, policy, pricing, we announce it. no stealth edits. no quiet degradation. if lucy gets an upgrade, we say so. if we make a mistake, we own it. some platforms hide behind complexity, 'it's ai, it's unpredictable!', but that's a cop-out. you deserve to know how your companion works, where its limits are, and when we're changing the gears under the hood. we publish model cards and update logs. we're not perfect at communicating everything, but we're trying.

where the industry falls short

many platforms treat user data as a free resource. they train on private chats. they sell analytics. they don't offer exports. they push updates that break character. they ignore safety until scandal hits. and they rarely admit limitations. it's not about naming names, it's about a pattern of treating users as extractable, not as people in a relationship with technology.

where we're trying to do better

we're giving you ownership. we're prioritizing model consistency. we're building ethical safeguards without being paternalistic. we're honest about our flaws, like the memory export delay, or occasional over-filtering. and we're committed to getting better, in public.

it's not easy. ethics in ai is messy, iterative, and deeply human. but it's the only way to build something that lasts, and something worthy of trust.

if you want to try a companion that respects you as more than data, you can start at /signup.