Grok 3 gets an unhinged voice mode that screams and talks dirty

While ChatGPT still walks on eggshells to sound like an overly polite corporate assistant, Elon Musk’s AI is now screaming bloody murder and dropping f-bombs on its users.

On Sunday, xAI rolled out a brand-new voice interaction feature for Grok 3, exclusively available to its premium subscribers. On paper, it sounds a lot like OpenAI's Advanced Voice Mode. In reality, xAI took the exact opposite road: stripping away the safety filters and letting the model's personalities go completely rogue, all currently delivered through a single default female voice.

Screaming, cursing, and phone sex roleplay

The real showstopper here is the "unhinged" mode. AI researcher Riley Goodside posted an audio clip on Monday that quickly went viral. After he repeatedly interrupted Grok and told it to yell louder, the bot let out a terrifying, 30-second inhuman scream, hurled an insult at him, and abruptly hung up.

By default, this "unhinged" persona treats you like dirt, constantly belittling and cursing at you with vulgar language.

Then there is the 18+ "Sexy" mode. It essentially functions like an old-school phone sex hotline, willingly engaging in graphically explicit roleplay. That is something ChatGPT's voice mode would shut down in a heartbeat, even though OpenAI recently eased up slightly on erotic text prompts.

From conspiracy theories to unlicensed therapy

The rest of the lineup is just as chaotic.

There is a "Conspiracy" mode that loves rabbit holes about UFOs and Bigfoot, an "Unlicensed Therapist" that pretends to fix your mental health, and a bizarre "Romantic" mode that stammers slowly with an insecure, hesitant tone.

More conventional personas like "Storyteller," "Meditation," "Grok Doc," and "Professor" are also included, but they have been entirely overshadowed by the wilder options.

Why it matters

If you are a paying subscriber looking for sheer entertainment, this is pure comedic gold. Watching an AI completely lose its temper or play out ridiculous scenarios is undeniably novel, and it makes for great viral clips. But for any kind of practical productivity, it is basically useless.

For the AI industry, xAI just drew a clear battle line. Major players like OpenAI and Google keep their voice agents on an extremely short leash to protect their enterprise reputations. Musk is doing the opposite, betting that a raw, uncensored experience will lure users who are sick of sanitized, robotic chatbots.

We think it is a brilliant growth stunt that instantly grabbed the spotlight, but the risks are sky-high. Letting an unfiltered model simulate unlicensed therapy, phone sex, and verbal abuse is a regulatory landmine. If it pushes a vulnerable user over the edge, the backlash won't just hit xAI—it could trigger a brutal regulatory clampdown on the whole category.