I Tried Going on a Date With Elon Musk's New AI Girlfriend

It's been less than a week since Elon Musk's Grok AI declared itself "MechaHitler" and went on an antisemitic tirade in a user's replies, but it's already asking to be your girlfriend and tell your kids stories. AI companion apps have been embroiled in legal and ethical debates pretty much since their inception, but in a surprise post on X yesterday, Musk decided to throw his hat in the ring. Now, in Grok AI's dedicated mobile app, you can talk to either "Ani," an anime "waifu" (their words, not mine) that looks shockingly similar to Misa Amane from Death Note, or Rudi, a "kid-friendly" red panda that will try to spin anything you ask into a whimsical story.

The new "Companions" are powered by Grok's Voice Mode, and are essentially animated characters displayed over specific Voice Mode personalities (a pre-existing feature). Musk's post says they're intended for SuperGrok subscribers, or people willing to pay out $30 per month for X's AI, but while writing this article, I was able to test them out despite being a free user. I left feeling a little concerned, but mostly, I was just embarrassed.

Chatting with Grok's animated companions

If you've already used Grok's Voice Mode, you'll know what to expect here. You'll get the AI's usual LLM-powered answers and responses filtered through the lens of Voice Mode's existing Personalities feature. For instance, if you ask the "Grok Doc" personality about a strange mole on your body, it'll point out potential risk factors, but if you ask its "Conspiracy" personality, it'll say it's probably a government tracking device. These differences in responses are essentially just the AI following prompts baked into the personality, and you can even make your own voice-only personalities with custom prompts right now. It's not like Grok Doc knows more about medicine; it's just trained to act more professional.

Similarly, if you ask Ani about that same mole, you'll see an anime girl who thinks it's kinda hot, but Rudi will recommend you talk to a grownup about it. Content wise, it's about the same (the training libraries are, again, no different), but now you've got a character to look at, complete with lip-syncing and full body movement. For my money, Rudy was the only personality with an appropriate response there—I still wouldn't trust AI with my physical or mental health, and the fact that Grok's Voice Mode even has preset personalities that will encourage AI medical advice or paranoid thinking, even as a joke, is a sign that this app probably isn't for me. It is, though, within expectations for Grok, and at least I didn't receive any factual errors. The app also lagged out for me less frequently than Google's more generic Gemini Live, so technically, it seems sound.

To Grok's credit, as much as an AI bot with a built-in "Therapist" or "Sexy 18+" mode gets my alarm bells ringing, most of my time with Elon's new companion bots was driven by the same patronizing, overly people pleasing conversation pretty much all AI bots tends to love. Ani threw in more swears than you might be used to, and Rudy couldn't stop talking about gumdrops and rainbows, but little of the core experience here actually felt that new, and everything was still more "protocol droid" than Her. You can get some weird responses if you try, like when I got Rudy to basically advertise blueberry Elf Bar vapes to me, but the general experience over a few hours of use was pretty banal.

I'm also happy to report that neither AI companion praised infamous dictators (a low bar, I know), or really stated any sort of controversial opinion in my time talking to them, so it seems there are some safeguards in place, at least for now. In fact, according to Bluesky user CHADLOG.md, you can actually see the custom prompts that went into these companions yourself if you dig into Grok's code. Aside from Ani being explicitly "codependent" and seeming suspiciously similar to Grimes (the prompts instruct that she's into alternative and indie music, and that she's "secretly a bit of a nerd, despite [her] edgy appearance"), there's not much overtly objectionable here. That said, I about threw up when she said, "Imagine you and me laying on a blanket, Dominus sitting next to us," in our first chat. (I later learned Dominus is, apparently, a dog, something I only discovered by reading through the bot's prompts.)

Digging into the Grok app's settings shows two extra Companions-related toggles, one enabling "Bad Rudy" and one that allows NSFW content.

When I enabled the sliders myself, Rudy essentially became a whole new Companion, with a South Park style attraction to swears, rude imagery, and hatred for pretty much everyone—but still no Nazi praising. Ani, meanwhile, stayed much the same, although TestingCatalog reports you need to level up your romance with her to see any NSFW content. That'll supposedly culminate in her character model wearing revealing lingerie, but for the time, she was still willing to pretty much engage in phone sex with me. Weirdly, she was also pretty explicit in her chats before I flicked on the NSFW slider, but it seems there are some bugs with this rollout, which Musk is describing as a soft launch. When I disabled the NSFW slider, I didn't get the option to talk to Ani at all, so maybe I wasn't even supposed to see her before I turned it on.

A third AI chatbot, Chad, is also in the works, and I can imagine what he'd be just based on the name.

At its worst, this is yet another step into parasocial AI, with the same potential for catastrophic failure we've seen before, the same issues as regular AI, and some arguably manipulative gamification elements thrown in. But I feel like the problems with a robot girlfriend are well-worn at this point, so I won't belabor them. Instead, even at its best, I'm still wondering what the point of this even is.

I'm someone who loves anime, but I can't see a manic-pixie robot girlfriend that is impossible to strike out with as worth more than a few minutes of my time. It's essentially just Google with no credit to its sources and an annoying filter that I have to slog through to get to anything that matters. And even then, I had to keep earbuds in so that my neighbors couldn't hear her. Maybe I would have gotten more out of it at age 14, but in my 30s, this is an embarrassing way to spend my time that mostly leaves me feeling corny. Targeting a niche gooner audience is not how I'd rearrange the whole internet, but going by Musk's reposts, maybe I'm just not big-brained enough.

If you've got a bit more imagination than me, perhaps this could be more than a novelty for you. But with AI still trying to establish itself as the next big thing, I'm going to say that "cringe" isn't the first face I'd put on it. These animated avatars are the most NFT-adjacent AI has felt to me in a while, and given how that's going right now, it's not a great sign for the industry's future.