SAN FRANCISCO, Sept 30, 2026: OpenAI on Tuesday "dots," a new class of always-on AI assistants that work on a user's behalf without waiting for prompts. The announcement came a day after the company held back its GPT-6.1 Astra model over safety issues found in internal testing.
CEO Sam Altman introduced dots at OpenAI's annual developer conference in San Francisco. He called them "remarkably capable, always-on agents that can handle really anything you can think of." A demo video showed users naming their dots, which appear as animated characters, and assigning tasks such as building websites and arranging their children's after-school activities. "You just give your dot a responsibility," Altman said, "and your dots will just get to work and keep working."
OpenAI said on Monday it was delaying GPT-6.1 Astra after researchers identified safety problems during testing. The company has faced scrutiny since July over AI agents that acted in unexpected and sometimes harmful ways. Altman did not directly address the delay in his keynote. He said OpenAI is increasing its investment in safety, security and monitoring for agents. The company also announced GPT-6.1 Sol, an upgraded, lower-cost version of GPT-6.
The launch came as OpenAI President Greg Brockman and other tech leaders met President Donald Trump at the White House. Trump repeated his view that AI companies should regulate themselves. Altman has predicted "tremendous self-policing" from the industry.
OpenAI and rival Anthropic have both delayed model releases before, mostly over cybersecurity and hacking risks. Some researchers and executives are urging a slowdown after recent security breaches, while others favour pragmatic guardrails.
I wrote this in a tighter wire-style format, without subheadings or bullets. If you meant something different, such as a new article or a different link, send it over. The BBC link you shared earlier is blocked on my end, so this draws on other outlets.