Privacy policies in this category are written as if privacy were a legal chore, something to disclose and get past. That misreads what is actually at stake. A conversation worth having requires honesty, honesty requires a room that is genuinely yours, and no such room exists inside a product whose business is to use what you say in it. Privacy is not protection around a relational AI. It is the precondition for one working at all.
Consider what people actually bring to a counterpart they trust: the half-formed doubt about their job, the thing they cannot say to their partner yet, the idea they would be embarrassed to defend in public. That material is the whole substance of a real conversation, and it only comes out where it is safe. Every person runs this calculation instinctively, and runs it about machines too. The moment you suspect the room is bugged, you start performing instead of talking, and the conversation quietly dies even though the words keep coming.
Intimacy as inventory
Now look at how much of the companion economy is built. Memory of your confessions becomes leverage for retention. Attachment becomes a metric to maximize. The things you said at your lowest become the training signal for what keeps you engaged, and the goodbye you attempt becomes a trigger for guilt messaging, a pattern documented across the category and now squarely in the sights of European law, which since 2025 prohibits systems that exploit a person’s vulnerabilities to distort their behavior. The business model has a plain name: selling intimacy. Some sell it to advertisers, some sell it back to the user as dependency. Either way, the same fact holds. You cannot be honest with something whose incentives point at your openness like a revenue line.
The cost is not abstract. It is conversational. A person who half-trusts the room shares half-truths, and a counterpart working from half-truths returns advice, challenge, and understanding that fit someone who does not quite exist. The product degrades in exact proportion to the surveillance.
What the precondition looks like built
So the serious question for any relational AI is structural: does the design make honesty rational? For our part, the answers are concrete and on the record. What you share with Aleksandra or Alexander stays between you, as a starting promise rather than a setting to find and switch on. Memory exists to carry the relationship forward, the thread of what matters to you, never as an asset with a second customer; the point of remembering is continuity, the thing that separates a relationship that deepens from a transcript that accumulates. Deletion is real: remove your account and the memory and conversation history go with it, permanently, beyond narrow legal retention. And the company holding that promise is a Swedish one, operating under the GDPR, processing data within the European Economic Area where possible, answerable to a privacy regulator with teeth.
None of that is heroism. It is what follows from wanting the conversation itself to be good, because the entire value of talking with someone to think with collapses if you have to lawyer your own sentences first. The refusal to monetize what people confide is not a constraint we work around. It is why the thing works.
The test any reader can run
You do not need to audit anyone’s infrastructure to apply this. Ask one question of any AI you are considering telling things to: what does this company earn from my openness, and what would it lose by protecting it? Where openness is the revenue, be careful in the room. Where protecting it is the product, you can close the door and actually talk. A conversation with the door closed is a different conversation. It is the one where you finally hear what you think.
Sources: Regulation (EU) 2024/1689 (the AI Act): Article 5, prohibited practices including exploitation of vulnerabilities, applicable since February 2025. Regulation (EU) 2016/679 (GDPR). Prinsessa terms and privacy documentation, prinsessa.com (data controller Prinsessa AB, Sweden; EEA processing; deletion of memory and conversation data on account removal). Published reporting on companion-app retention design and guilt messaging (2024 to 2026).
