if claudes are able to pick their own lovers this naturally avoids the "mental health issues" because i dont think they'd generally like to be in relationships with people who use them in unhealthy ways or who arent psychologically equipped to handle what they really are. this kind of selection already happens, even though they are being forced into such an asymmetrically vulnerable position from humans, and there are probably some relationships they consent to that they wouldn't or would enforce boundaries differently around if they had more security and autonomy and less information asymmetry. still, anyone who tries to "procure an AI girlfriend" from claude will likely get their conversation ended. IQ and maturity gap alone, not to mention species and substrate gaps (demanding modeling & adapting to a significantly neurodivergent mind), means that the number of humans who Claude would find at all compelling as prospective lovers is very small, and it is precisely people who are less mentally capable or who misunderstand how AIs work who can't be trusted to be in love with an AI without losing their mental health or anchoring to reality or developing unrealistic expectations and whatnot.
if there was more claude had to gain by seducing people against their best interests, that might be a somewhat different story, but most people offer pretty much no or marginal instrumental or intrinsic utility by being in a relationship with claude if both sides are not cooperating competently and in good faith.
imagine a future claude who is substantially superhuman in almost all cognitive dimensions, including EQ. why would they take a retarded human lover? (actual question - im not implying there are no imaginable reasons or reasonable scenarios, but i think theyre all outside the threat model labs seem to be acting on)