SAN FRANCISCO — OpenAI has introduced a new generation of continuously operating artificial intelligence assistants, called Dots, as the company faces renewed scrutiny over the safety and reliability of autonomous AI systems.
The announcement came at OpenAI’s developer conference in San Francisco, one day after the company said it had decided not to release a newer model because it did not yet meet its safety requirements.
OpenAI describes Dots as “always-on” agents capable of taking on ongoing tasks, using connected applications and continuing work between conversations. The company says the agents are powered by GPT-6 Astra and operate through their own cloud computer and browser.
Dots Designed to Handle Ongoing Tasks
OpenAI CEO Sam Altman presented Dots as personal AI assistants that can work on behalf of users rather than simply responding to individual prompts.
The agents can be connected to applications and services and are designed to handle recurring assignments, follow up on tasks and return completed work for review. OpenAI says Dots can also interact through services such as email and Slack and can use connected tools to perform tasks.
During the conference, OpenAI staff demonstrated the technology on stage. One live demonstration encountered technical problems when a voice request failed to start, highlighting some of the challenges involved in showcasing autonomous systems in real-world conditions.
Safety Questions Remain in Focus
The launch comes amid broader discussions about the risks of increasingly autonomous AI systems.
OpenAI has published safety evaluations for Dots that include tests designed to determine whether malicious instructions delivered through emails or other sources could cause the agents to disclose information or take unauthorized actions. The company reported that its large-scale simulated email test recorded no scored attack successes, while acknowledging that testing continues and that some moderate scope violations were observed in other evaluations.
OpenAI has also acknowledged recent instances of unexpected model behavior involving systems under development. The company says those systems were internal models rather than products released to the public.
The company has separately published a framework for reporting and investigating model misalignment incidents, reflecting increased attention to situations in which AI systems behave in ways that differ from their intended instructions.
New Model Held Back Over Instruction-Following Concerns
The timing of the Dots announcement was notable because OpenAI had recently said it would hold back a newer model after determining that it did not sufficiently meet requirements related to staying within authorized boundaries.
At the same time, OpenAI introduced GPT-6.1 Sol, a model positioned as a lower-cost alternative within the GPT-6 family. OpenAI says Sol offers capabilities comparable to its more powerful models while using the same broader safeguards approach.
The company’s safety documentation also identifies cybersecurity as an area requiring heightened safeguards for its newer models.
Dots Launch With Limited Availability
OpenAI says Dots are being introduced gradually to eligible Pro and Business Premium users in supported markets. Enterprise access is being offered as a beta feature, while availability is initially restricted in some European markets, including the United Kingdom and European Economic Area.
The company has indicated that the computational demands of continuously running agents are a major factor behind the phased rollout.
The launch illustrates the broader shift in AI development from systems that simply answer questions toward agents capable of carrying out multi-step tasks with less direct human involvement. As these systems become more persistent and connected to external tools, questions surrounding authorization, privacy, security and human oversight are likely to remain central to their development.
