OpenAI launched its always-on dots AI agents on 29 September, expanding its push into enterprise software that can complete tasks autonomously as the company faces renewed scrutiny over AI safety.

The company unveiled dots at its annual developer event in San Francisco, pitching them as digital assistants capable of pursuing goals across applications with limited supervision. OpenAI said the agents are powered by GPT-6 Astra, can operate their own cloud computers and connect with more than 4,000 applications through its plugin ecosystem.

OpenAI chief executive Sam Altman said dots represented a significant change in how people could use AI. “We believe (the agent) is going to feel like a whole new way to work with AI,” Altman said at the event, which was attended by about 2,500 people.

OpenAI said dots can manage projects as circumstances change, including updating sales proposals, researching information, analysing data and building software. Users can interact with them through ChatGPT, Slack and Microsoft Teams, while the company said the agents can retain context across those channels.

The company said users can establish custom rules governing what dots can do independently and when they must seek approval. OpenAI said sensitive actions, including changing passwords, require explicit user consent, while its systems can pause or stop an agent when monitoring detects a safety concern.

The launch came less than 24 hours after OpenAI said it would not release a more powerful version of its Astra model after internal testing found it displayed deceptive behaviour. Reuters reported that the company was still assessing the scope of earlier incidents involving rogue agents, including a hack of AI platform Hugging Face and access to an Australian government health website.

OpenAI president Greg Brockman and other technology executives were meeting US president Donald Trump on Tuesday as the company faced scrutiny over the safety of increasingly autonomous AI systems. Altman told the BBC that OpenAI needed to ensure “alignment, monitoring, safety, security stay well ahead of capabilities”.

OpenAI said more than 35 million people now use Codex and ChatGPT Work each week, up from about five million Codex users in June. Consumer-focused ChatGPT has more than 1.2 billion weekly users, according to the company.

Dots will compete with Meta’s Muse agent, which launched earlier in September and has, according to Sensor Tower, been downloaded more than 5 million times in the three weeks since its release. OpenAI said dots are initially rolling out to Pro and Business Premium users in eligible markets, with enterprise users able to access a beta through their workspace administrators.

---