OpenAI has turned ChatGPT's voice feature into something you can hand real tasks to, instead of just a way to talk to a chatbot. The update lets you speak to ChatGPT and have it check your calendar, send an email, read through Slack, or catch a duplicate charge in a finance app and cancel it. It runs on three new AI models, a large one called Astra and two smaller, cheaper ones called Sol and Luna, all released this month.
The more significant change is where voice now works. It has been added to "ChatGPT Work," OpenAI's tool for building documents, spreadsheets, presentations, and simple websites. Until now, getting that kind of output meant typing instructions and waiting. Now you can talk through a task on your phone while doing something else, and the finished document, slide deck, or site shows up when you check back.
This is not a small feature update tucked into a release note. It marks OpenAI's clearest move yet toward an assistant that takes action on your behalf rather than one that answers questions. Sam Altman has used the movie "Her" as his reference point for this vision since 2024, describing an AI you talk to naturally and that quietly runs parts of your life. That comparison has not been free of trouble: OpenAI pulled one of its voice options in 2024 after it sounded close enough to actress Scarlett Johansson that she took legal action.
The timing also puts OpenAI in a race that is now visibly two-sided. Anthropic, the company behind Claude, merged its own chat and work tools into a single interface just a week before this OpenAI release, letting users create documents and slides without switching tabs. OpenAI is taking a different structural approach, keeping voice, chat, and Work as separate tabs rather than one blended screen, but the direction both companies are moving in is the same: voice and action are becoming one feature instead of two.
For a business, the interesting part is not how natural the voice sounds. It is what you are agreeing to when an employee gives a voice assistant standing access to email, calendar, and internal chat. A typed command leaves a paper trail you can review before it runs. A spoken command that triggers an action immediately does not always give you that same pause. Companies that already limit which apps and data an AI assistant can touch in text form need to ask the same question about voice, because right now most have not set a policy for it at all.
ChatGPT already has an enormous reach, with weekly users in the hundreds of millions, so a change like this does not stay a niche feature for long. If voice becomes the default way people interact with ChatGPT at work, the permissions attached to that voice channel deserve the same scrutiny as any other piece of software that touches your inbox, your calendar, and your company's internal messages.