OpenAI’s new voice mode makes it to the ChatGPT desktop app
OpenAI has added its new ChatGPT Voice capability to the ChatGPT desktop app, enabling people to speak commands that direct AI agents and carry out tasks on their computers. The change matters because it moves the feature beyond conversational voice interaction towards hands-free, multi-step computer assistance for work and coding.
The feature uses OpenAI’s recently launched ChatGPT-Live voice models and works with ChatGPT Work and Codex, including computer-use tools for websites and apps. On macOS, Appshots can give the app access to on-screen content, including alt text; OpenAI demonstrated a developer using one spoken instruction to create a thread, open a pull request and investigate a bug. The rollout began globally on 23 July, while iOS users can access Codex voice remotely but cannot have the voice feature take actions directly on their phones.
In short
- ChatGPT Voice now controls desktop AI agents through spoken commands.
- It supports multi-step tasks in ChatGPT Work and Codex.
- macOS users can optionally share on-screen content with the app.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
The integration of voice control into desktop AI assistants represents a valuable advancement in human-computer interaction that enhances accessibility for users with mobility limitations and streamlines workflows for knowledge workers. This capability allows complex multi-step tasks to be completed more efficiently through natural language commands, and the feature operates within a user's own system with explicit consent, relying on existing computer protections. As demonstrated, developers can accomplish in a single spoken instruction what previously required multiple manual steps, representing genuine progress in productivity.
The case against
Granting AI systems direct execution authority over computer operations introduces meaningful security and privacy risks that warrant careful consideration. Even with good intentions, automated systems can misinterpret voice commands, encounter edge cases their training did not anticipate, or become vectors for exploitation if compromised. Users may not fully appreciate the implications of permitting an AI agent autonomous control over their system, and the potential for unintended consequences—data access, system changes, exposure of sensitive information—demands robust safeguards, transparency about limitations, and deliberate implementation rather than rapid rollout.
Explained simply
ChatGPT is an AI tool created by OpenAI that people use to get answers, help with work and solve problems. Most people interact with it by typing, but OpenAI has been developing voice capabilities that allow users to speak to the AI instead.
Voice assistants have traditionally focused on answering questions, but newer AI systems can do more when instructed through voice: they can carry out tasks on computers, access on-screen information and direct software tools. This is particularly useful for technical work where managing multiple applications simultaneously is common.
As these voice capabilities have improved, OpenAI has expanded their availability across different platforms and versions of ChatGPT. This reflects a broader trend of AI tools becoming more deeply integrated into work routines, especially in professional settings where being able to operate tools hands-free could save time and improve efficiency.