OpenAI Unveils 'Always-On' AI Agents a Day After Shelving Model

OpenAI CEO Sam Altman introduced a "remarkably capable, always-on" artificial intelligence agent on Tuesday, one day after the company halted the rollout of a more advanced model over safety concerns raised by its own researchers.

The announcement came at OpenAI's annual developer conference, where Altman presented a string of product launches and updates headlined by OpenAI's new agents, called Dots. The agents are designed to complete ongoing tasks proactively on behalf of users. They position OpenAI in direct competition with Meta's personal agent Muse, which exploded in popularity after Meta's own conference last week.

Dot agents will be "like an AI helper that always has your back, inspired by the cool versions of what we all watched in movies growing up," Altman said.

A launch shadowed by a safety delay

The timing was awkward. On Monday, OpenAI said it was holding off on releasing the other model because of concerns raised by its researchers. That decision followed broader calls within the industry to decelerate the pace of the technology's advance and let safety measures catch up.

Altman avoided referencing the shelved model during his keynote address. But during a question-and-answer session, he said the company was investing more in safety, security and monitoring of AI agents. (An earlier version of the story did not reflect that the security discussion occurred during the Q&A rather than the keynote.)

Altman framed the AI boom in unusually humanistic terms, arguing it should be considered more like a period of renaissance than an industrial revolution.

"The best version of AI is not about making people cogs in a giant machine, ever whirring faster and faster," Altman said. "There are some parts of life that we cannot and should not automate. AI should be about giving people more power over their own lives, more tools to create, learn, discover, expand knowledge."

What else shipped

Beyond the Dots, OpenAI announced GPT-6.1 Sol, an upgraded version of its model GPT-6 Sol. The company also launched a premium speed tier called "Ultrafast," among other releases and updates.

The product cadence shows OpenAI pressing ahead on commercial expansion even as internal safety deliberations delayed one of its most anticipated models. The gap between the two events — a shelved release on Monday, a festival of launches on Tuesday — captures the central tension now facing the industry: how quickly to ship versus how carefully to test.

CEOs sign self-policing accord at the White House

The announcements landed just before a separate, high-stakes event in Washington. On Tuesday, leaders from six of the leading AI companies, including OpenAI, signed a voluntary accord with President Donald Trump to "self-police" development.

Under the accord, the companies committed to implementing "robust internal controls," partnering with an "independent external auditor" to assess whether those controls were working, and setting up a committee among each company's board of directors to evaluate reports from internal and external auditors.

Altman did not attend. OpenAI president and co-founder Greg Brockman represented the company at the White House, alongside Dario Amodei of Anthropic, Amazon chief Jeff Bezos, Nvidia's Jensen Huang, Tesla and SpaceX's Elon Musk, and Microsoft CEO Satya Nadella.

The accord reflects a sharp divide over how AI should be governed. Many in Silicon Valley have called for the government to help establish guardrails for the technology. Trump has pushed back against the idea of greater government oversight of AI, and the voluntary deal keeps enforcement largely in the companies' own hands.

The stakes

Taken together, the week's events sketch the industry's current balancing act. OpenAI delayed a frontier model on researcher concerns, then shipped an agent platform designed to embed its technology deeper into users' daily workflows. Six of the largest AI players pledged to audit themselves, with no government enforcer in the loop.

For OpenAI, the Dots launch also answers a competitive threat. Meta's Muse demonstrated last week how quickly a consumer-facing agent can capture attention, and OpenAI's answer arrives with a familiar pitch: proactive help, always available, modeled on the science-fiction assistants of earlier decades.

Whether the industry's new internal controls and external audits satisfy the researchers calling for slower development — the same concerns that grounded OpenAI's more advanced model — will shape how quickly the next generation of agents reaches users.