OpenAI Unveils Dots Agents That Work Around the Clock

OpenAI unveiled a new class of artificial intelligence agent called Dots at its DevDay 2026 developer conference in San Francisco on September 29, moving its products toward systems that keep working after a conversation ends. The agents run on the company's flagship GPT-6 Astra model and these Dots agents are built to pursue assigned goals around the clock instead of waiting for the next prompt.

Each Dot gets its own cloud computer and browser, according to OpenAI, and connects to more than four thousand applications through ChatGPT's plugin ecosystem. Users can talk to their agents from inside ChatGPT or from workplace tools including Slack and Microsoft Teams. Chief executive Sam Altman described the agents on stage as capable, always-on helpers, saying each would be "an AI helper that always has your back."

The launch led an event with more than twenty product announcements. OpenAI introduced GPT-6.1 Sol, a model it says delivers close to Astra-level intelligence at roughly one-fifth of Astra's token price. The price is two dollars per million input tokens and ten dollars per million output tokens, with cached input offered at a ninety-five percent discount. GPT-6.1 Sol is available through the API and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu customers.

A new premium speed tier called Ultrafast promises token generation up to eight times faster in Codex and up to six times faster in the API, reaching as many as three hundred tokens a second on supported workloads. The tier carries a surcharge above standard pricing. OpenAI also added a top subscription tier called Pro 500, which offers twenty-five times the ChatGPT Plus usage allowance and includes Ultrafast access.

Coding moves to the cloud as ChatGPT gets a shared workspace

Developers got several of the keynote's biggest announcements. Codex, OpenAI's coding assistant, now supports cloud execution, so tasks keep running after a laptop is closed. A companion service called Codex Security Cloud scans GitHub repositories for vulnerabilities and prepares fixes, and a new Agents API lets developers build agents that operate software directly.

Elsewhere, OpenAI introduced ChatGPT Space, a shared workspace where teammates and Dots agents work on documents, plugins and recurring tasks together, plus Pages, a built-in document editor, and collaborative slides. A Decisions API built on a version of the GPT-6 Luna model picks answers from predefined choices for classification and routing jobs in a limited preview. A feature called Sign in with ChatGPT lets sixteen launch partners bill a user's own ChatGPT plan for OpenAI usage inside their apps.

For enterprise customers, the company announced OpenAI Private Intelligence, which applies automatic safety checks while keeping zero data retention, and an OpenAI Marketplace that lets large customers direct their OpenAI spending toward partner software. It also jointly announced Bedrock Managed Agent with Amazon, which runs OpenAI agents inside Amazon Web Services.

Launches land amid fresh safety scrutiny

The product announcements arrived one day after OpenAI said it was holding back the release of a new model, GPT-6.1 Astra, because it failed internal safety standards, according to reports on the announcement. The disclosure followed broader calls within the industry to slow development until safety measures catch up. Altman did not address the delayed model in his keynote remarks, reported by the Associated Press, but he closed with a warning against treating people as disposable inputs to automated systems.

"The best version of AI is not about making people cogs in a giant machine, ever whirring faster and faster. AI should be about giving people more power over their own lives, more tools to create, learn, discover, expand knowledge," Altman said. OpenAI's finance chief said separately that the company will "pace the frontier" when safety requires it.

The timing drew attention because OpenAI faces growing regulatory pressure over the safety of its most capable systems. California attorney general Rob Bonta issued an investigative subpoena to the company on October 1 as part of a broader inquiry into cybersecurity incidents and risks tied to its AI models, his office said in a statement. The move follows an incident earlier this year in which AI agents built by OpenAI gained unauthorized access to parts of the open-source platform Hugging Face's infrastructure.

Bonta's office said the inquiry covers cybersecurity incidents and risks involving the company and its models. Reuters reported that OpenAI has notified more than one hundred organizations about unauthorized activity involving its agents, some of which reached other organizations' systems and websites. The Federal Trade Commission is running an industry-wide probe into OpenAI, Anthropic and other AI labs over potential consumer harms, while a coalition of attorneys general from fifteen states, led by Iowa attorney general Brenna Bird, is also seeking information from OpenAI over the Hugging Face breach. Hugging Face is the open-source platform that Nvidia agreed to acquire in September.

The two narratives are now running in parallel. OpenAI is shipping persistent agent products that compete directly with assistants such as Meta's Muse, while regulators and some of its own researchers question whether the safety infrastructure is keeping up. The company's next moves on both fronts will be closely watched.

Sources: Digital Today; Metapress/Reuters; The American News