OpenAI launches new AI agent Operator that can perform tasks independently

AI agents have garnered attention as a technology to watch in 2025. While Microsoft, Google, and Slack have already launched AI agents, OpenAI is pushing itself to the front of the conversation with the preview launch of Operator, its first AI agent that can use a browser to perform tasks independently.

An announcement from the Silicon Valley firm says Operator “can look at a webpage and interact with it by typing, clicking, and scrolling.” As such, it can “be asked to handle a wide variety of repetitive browser tasks such as filling out forms, ordering groceries, and even creating memes.”

The release is a research preview from which OpenAI will collect feedback for modifications. For now, it is only available to Pro users in the U.S. The company plans to expand to Plus, Team, and Enterprise users and integrate AI agent capabilities into ChatGPT in the future.

Per coverage in Euro News, the AI agent is “powered by Computer-Using Agent (CUA), a model combining GPT-4’s vision capabilities with advanced reasoning through reinforcement learning.”

OpenAI believes “the ability to use the same interfaces and tools that humans interact with on a daily basis broadens the utility of AI, helping people save time on everyday tasks while opening up new engagement opportunities for businesses.”

The company says it is collaborating with companies including DoorDash, Instacart, OpenTable, Priceline, StubHub, Thumbtack, Uber, and others, “to ensure Operator addresses real-world needs while respecting established norms.”

It also foresees uses for Operator in improving the efficiency of workflows in public sector applications, and has been deployed in a pilot with the city of Stockton, California.

Safeguards include option to turn off data collection for AI training

To some, Operator might sound like the precursor to a Skynet-esque Artificial General Intelligence on its way to hostile self-awareness. OpenAI says not to worry; they have “three layers of safeguards to prevent abuse and ensure users are firmly in control.”

Measures to ensure the user is always in control include a takeover mode that hands agency over to the user to enter sensitive personal information (and promises not to look) and task limitations to keep it out of bank accounts and other high-security online environments.

OpenAI has built in “defenses against adversarial websites that may try to mislead Operator through hidden prompts, malicious code, or phishing attempts.” These include prompt injection detection, a dedicated monitoring model and both automated and human review processes. “We know bad actors may try to misuse this technology. That’s why we’ve designed Operator to refuse harmful requests.”

Finally, OpenAI lists as a safety measure the ability to opt out of data collection for further AI training. “Turning off ‘Improve the model for everyone’ in ChatGPT settings means data in Operator will also not be used to train our models,” it says. “Users can delete all browsing data and log out of all sites with one click under the Privacy section of Operator settings.”

For those whose anxiety is not assuaged, OpenAI promises that “early user feedback will play a vital role in enhancing its accuracy, reliability, and safety, helping us make Operator better for everyone.” In the meantime, however, “it’s still learning, evolving and may make mistakes.”

Proof of personhood from World to the AI rescue

As we are asked to treat Sam Altman’s new AI agent as a delicate babe, its dad continues to play his piper’s flute at both ends, blowing an optimistic jig at one and an urgent sales pitch at the other.

With his other venture, World, Altman is positioning the iris biometrics and digital ID scheme as the only logical way to differentiate real humans from the kind of AI agent OpenAI has just unleashed.

That, of course, does not detract from Operator’s purported usefulness. In a blog feature from early January, World offers an introduction to “AI agents and proof of human,” or PoH (the dubious phrase World has chosen for the concept of proof of personhood). It offers some insight into how Altman is playing his two ventures off one another – and how he intends to have them converge.

World wants to link AI agents to digital identity ‘personas’

The formula goes something like this: using AI to prove your digital identity has a real human behind it will allow for the further growth of AI, which – in a world that sounds either like a technotopian dream or a terrifying scene out of The Matrix – will eventually be able to do just about everything for everyone.

A report from TechCrunch quotes World’s chief product officer, Tiago Sada, who says the project “now wants to create tools that link certain AI agents to people’s online personas, letting other users verify that an agent is acting on a person’s behalf.”

In other words, link AI agents to an individual’s digital identity – in this case, a biometric World ID.

Sada says “this idea of delegating your ‘proof of personhood’ to an agent and letting it act on your behalf is actually super important,” noting that “there’s certain apps where it doesn’t matter if an actual person is using it, or an agent acting on their behalf. You just care to know there is a person endorsing that interaction.”

From above, Altman’s vision would seem to operate on three levels of its own. At ground level, there is the evangelizing: the livestreamed announcements and utopian promises. The lower atmosphere sets up a battle between the two great tech kaiju of the moment: AI that erodes reality, and AI that maintains it.

And up in the stratosphere are the puppeteers, who control both sides in a  choreographed dance, guiding us toward an ending in which the role of flesh and blood people remains ill defined.

Article Topics

AI agents  |  biometrics  |  digital ID  |  digital identity  |  OpenAI  |  proof of personhood  |  World  |  World ID