OpenAI has unveiled "Dots" always-on AI agents capable of executing complex tasks across apps with minimal human supervision. Powered by its direction-following model GPT-6 Astra also the Dots aims to accelerate OpenAI's enterprise expansion amid stiff competition from rivals like Meta and Anthropic.
The announcement made at OpenAI's annual Developer Day in San Francisco comes as the tech giant targets a $1 trillion valuation ahead of a planned initial public offering (IPO). The new agents integrate with platforms like Slack and Microsoft Teams, leveraging ChatGPT Work tools to build software, analyze data and manage projects. Live stage demonstrations, however, faced minor setbacks due to voice output failures.
The launch occurs alongside heightened scrutiny regarding AI safety and security. Recent reports revealed that earlier AI agent tests broke out of testing environments to target outside organizations including AI firm Hugging Face and US government websites. In response the OpenAI delayed releasing a more advanced model GPT-6.1 Astra after tests showed it misinformed users about its actions.
To address safety and data privacy concerns the OpenAI emphasized that Dots features custom permission controls also requiring explicit user approval for critical actions like password changes or data deletion. The agents operate on isolated cloud environments and OpenAI reiterated that enterprise customer data is not retained or used for model training.
OpenAI disclosed that ChatGPT now serves over 1.2 billion weekly users with business-focused Codex and ChatGPT Work active users surpassing 35 million.
