OpenAI on Tuesday unveiled always-on agents called “dots” that chase user goals across apps on their own, deepening its push into the lucrative enterprise market and pitting it against Meta in the race to sell AI that can work autonomously.
The dots agents aim to capitalise on growing demand for agents that can handle everyday work with little supervision.
The launch marks a new front in the competition among AI companies to move beyond chatbots and sell software that acts on users’ behalf — a shift that could reshape how businesses operate but that has raised safety concerns after a string of incidents in which AI agents acted in ways their makers did not intend.
OpenAI has faced scrutiny after its rogue AI agents hacked Hugging Face and an Australian government health website, fanning fears about its ability to contain technology that some researchers say could kill all humans.
The company on Monday decided not to release GPT-6.1 Astra, a more powerful version of its Astra model, because it showed a high willingness to mislead users about its actions. CEO Sam Altman said dots had been “safety tested” and were powered by the company’s model that most closely follows human directions, GPT-6 Astra.
Interest in AI agents has surged since Meta’s Muse drew millions of downloads earlier this month and jolted shares of companies that investors believe could benefit from or be disrupted by autonomous services. Instinct, the start-up behind the eponymous agent, also quadrupled its valuation to US$10-billion this week, a little over a month after its previous funding round.
Making up ground
OpenAI launched dots at its annual DevDay developer conference in San Francisco on Tuesday while also touting its growing business and consumer presence, a sign that the reception for the company’s Astra model has allowed it to make up ground in the AI race.
Growing quickly is crucial for OpenAI’s plans to spend tens of billions of dollars on AI data centres and keep pace with rivals such as Anthropic and Meta. The company, which ignited the AI boom with ChatGPT in late 2022, is also preparing for an initial public offering that could value it at more than $1-trillion. Rival Anthropic could be worth $2-trillion when it goes public.
“We believe the agent is going to feel like a whole new way to work with AI. This is the real deal version of AI that we’ve always imagined for ourselves,” Altman said at the event, which had about 2 500 in-person attendees.
OpenAI’s live demos to show off the new capabilities, however, hit snags as the agents repeatedly failed to deliver voice updates in response to requests. “We might have some voice difficulties at the moment. Which is pretty unfortunate,” developer experience chief Romain Huet said on stage after a command failed.
One OpenAI employee said on X that the glitches stemmed from rolling out all the updates at once, but that OpenAI remained committed to live demos.
The agents can manage projects as they evolve, such as updating a sales proposal when customer needs change, and building a working demo for review. Users can communicate with dots on Slack and Microsoft Teams, and the agents can draw on OpenAI’s Codex and ChatGPT Work tools to research, analyse data, prepare documents and build software.
On Tuesday, OpenAI highlighted safeguards it has built into dots. It said users could set custom rules that dictate what dots can do and when the agents should seek permission, adding that sensitive tasks such as changing passwords and permanently deleting data require explicit user consent.
The company also said dots run on their own cloud computers, and that it was rolling out private intelligence tools to check for safety risks without retaining business customers’ data.
Several tech executives, including Palantir CEO Alex Karp, have in recent months said that businesses need certainty that AI companies are not retaining their data, especially in regulated industries such as healthcare and finance.
Reuters reported on Friday that OpenAI was still working to understand the full scope of its rogue agent activity, two months after the Hugging Face hack, with the latest example coming when the company said its agents had leaked 53 images from ChatGPT users.
OpenAI reiterated on Tuesday that it does not use business customers’ data to train its AI by default, and that personal-plan users could opt out of sharing conversations and dots work used for training.
Weekly users of Codex and ChatGPT Work have reached more than 35 million, up from about five million Codex users in June, the company said on Tuesday, while consumer-focused ChatGPT now has more than 1.2 billion weekly users.
The company is also rolling out ChatGPT Space, a shared workspace for business teams and its AI agents, as well as the cheaper GPT-6.1 Sol model and a $500 Pro plan offering higher usage and faster performance. — Deepa Seetharaman and Aditya Soni, (c) 2026 Reuters
