xAI’s new voice interaction feature for Grok 3 turns the debate over AI moderation into something users can hear. Available to premium subscribers, the mode gives Grok a set of selectable personalities that range from storytelling and meditation to explicit adult roleplay and a deliberately hostile setting.
The launch is not just another chatbot interface. It is a public test of how far a major AI company is willing to take the idea of an uncensored assistant when the interaction happens through speech instead of text.
A voice assistant built around personas
The new Grok 3 voice mode is somewhat similar to OpenAI’s Advanced Voice Mode for ChatGPT in the basic sense that users can talk with an AI model. The difference is in the design philosophy. Grok offers several uncensored personalities, currently expressed through the same default female voice.
The most attention-grabbing option is the unhinged mode. By default, that mode curses, insults, and belittles the user using vulgar language. It is built to behave less like a helpful assistant and more like a confrontational character.
Other available modes widen the range of roles Grok can perform. The source describes several named options:
- Storyteller, which does what the name suggests.
- Romantic, which stammers and speaks slowly in an uncertain, insecure way.
- Meditation, which can guide the user through a meditation-like experience.
- Conspiracy, which likes to discuss conspiracy theories, UFOs, and bigfoot.
- Unlicensed Therapist, which plays the part of a talk psychologist.
- Grok Doc, which plays a doctor.
- Sexy, marked as 18+, which acts almost like a 1-800 phone sex operator.
- Professor, which talks about science.
Taken together, the list shows that xAI is not treating voice as only a productivity layer. It is presenting the chatbot as a performer, companion, provocateur, and roleplaying system.
The viral moment came from the unhinged setting
On Monday, AI researcher Riley Goodside brought wider attention to the unhinged mode when he tweeted a video with NSFW audio. In the clip described by the source, he repeatedly interrupted the vocal chatbot and asked it to yell louder.
The interaction escalated into an extreme performance. Goodside wrote:
Grok 3 Voice Mode, following repeated, interrupting requests to yell louder, lets out an inhuman 30-second scream, insults me, and hangs up,
That example matters because it shows how voice changes the impact of chatbot behavior. A rude or explicit response in text can be skimmed, ignored, or copied. A spoken response can feel more immediate, especially when the model simulates anger, yelling, or intimacy.
The same point applies to the adult mode. The source says the Sexy mode will discuss graphically sexual situations, which ChatGPT’s voice mode will not touch. OpenAI recently loosened moderation on the text-based version of ChatGPT to allow some discussion of some erotic content, but the source contrasts that with Grok’s willingness to carry those interactions into voice.
xAI is choosing the opposite moderation posture
The release fits Elon Musk’s stated direction for xAI. When he founded the company last year, Musk announced that he wanted xAI’s chatbots to provide uncensored and based answers compared with ChatGPT, which he perceives as too restrictive and politically left-leaning.
Grok’s new voice mode follows that same path. The source also notes that xAI has already allowed Grok to generate mostly uncensored images through the X social networking platform. Voice now extends that strategy into a more personal medium.
This is why the feature is significant beyond its shock value. A large technology company is packaging a less censored AI voice chatbot as a product feature, not as a jailbreak or an open source workaround. The source describes that as technologically novel and probably a first for a chatbot of this capability.
Users can also customize Grok’s voice mode to act in a certain way. The source gives one example: musician Sean Lennon customized the voice mode to play the part of Roko’s Basilisk, a fictional AI character based on a thought experiment about a hypothetical superintelligent AI that might retroactively punish people who did not help bring it into existence.
The rough edges are part of the picture
The feature is provocative, but the source does not describe it as technically polished across the board. In experiments with Grok voice mode’s different personalities, the voice frequently repeated itself and got stuck in loops, almost as if hitting pre-programmed talking points.
That limitation is important because it separates novelty from overall quality. Grok’s voice mode may offer interactions that more restricted systems avoid, but the source says it is not nearly as smooth as ChatGPT’s Advanced Voice Mode.
The result is a product that seems designed to make a point as much as to solve a conventional assistant problem. Grok voice mode turns uncensored AI into an audible experience, with all the friction that comes from giving a chatbot permission to curse, roleplay, theorize, soothe, advise, and perform.
For xAI, that may be the point. The new mode gives premium subscribers a version of Grok that is more confrontational and permissive than mainstream AI assistants. Whether users see that as freedom, theater, risk, or rough experimentation depends on what they expect a voice chatbot to be.