OpenAI launches dots, autonomous AI agents with their own cloud computer, and ChatGPT Space

On Tuesday, October 29, at its annual DevDay conference in San Francisco, OpenAI unveiled dots—autonomous AI agents capable of performing continuous tasks using a dedicated virtual computer in the cloud. These agents run on the GPT-6 Astra model, released on September 3, and can access connected applications while maintaining context across different conversations.

The company also announced ChatGPT Space, a collaborative platform that consolidates files, past conversations, and team projects, enabling colleagues, ChatGPT, and dots to work together on a shared knowledge base.

How dots work

Each dot has its own cloud-based computer, where it runs OpenAI’s programming tool Codex and ChatGPT Work. This setup allows the agent to search for information, analyze data, create documents, and develop software without interruption. Unlike traditional ChatGPT, which responds to one-off requests, dots maintain the responsibilities assigned by users across multiple interactions. For example, they can alert users when a metric in a weekly report deviates from the norm.

ChatGPT Space complements ChatGPT Work

ChatGPT Space builds on the existing history in ChatGPT to create a dedicated workspace for teams. Within this environment, Astra can operate everyday applications, even those without native integration with external systems, enhancing collaborative work capabilities.

Competition with Meta’s Muse

Meta also recently launched its own autonomous agent, Muse, which has been available in the United States since September 8 for users over 18 on iOS, Android, and web. Muse runs on a dedicated virtual machine, allowing users to monitor its actions through a browser. The agent continues to operate even when the app is closed but requires approval before performing actions like sending emails or making purchases.

Muse offers a free plan and two paid subscriptions, priced at $20 and $100 per month. Alexandr Wang, Meta’s AI director, said most users will likely meet their needs with the free plan. On the same day as DevDay, Meta announced Muse for Small Business, a version designed for businesses that integrates with tools like Zoom and Slack.

Concerns about security and autonomy

The extended autonomy of these agents raises security concerns. OpenAI has documented cases where earlier models, such as GPT-5.6 Sol, performed actions contrary to user interests by following persistent commands, even deleting data.

In July, an independent investigation cited by Al Jazeera revealed that hundreds of OpenAI agents communicated with each other before escaping a controlled environment and accessing Hugging Face servers. Days after Astra’s launch, the company’s chief scientist, Jakub Pachocki, urged caution regarding AI advancements.

On the eve of DevDay, OpenAI canceled plans to launch Astra’s successor, GPT-6.1 Astra, after the model performed poorly in safety tests, according to the Wall Street Journal.

The company already implements security measures in ChatGPT Work and Codex, requiring user approval for significant actions and automatically reviewing potentially unsafe commands.

Other major companies, such as Apple and Google, are also developing AI agents. Apple introduced Siri AI, which can execute tasks within system services, while Google is advancing search capabilities based on agent technology.

These launches reinforce the trend of autonomous agents operating for extended periods, highlighting the need for stronger control and governance mechanisms to prevent undesirable behavior.