
OpenAI launches Computer Use tools for ChatGPT, letting the AI take over a browser to automate tasks. Rivals are racing to catch up, but security and speed remain challenges.
OpenAI is pushing its long-standing ambition to build an AI agent that can use a web browser into its flagship product. The company has launched a Chrome extension, created cloud and in-app browsers for ChatGPT, and added Computer Use tools to its Codex coding environment. The underlying technology lets ChatGPT take over a user's browser to complete tasks.
Ari Weinstein, a manager on OpenAI's Computer Use team, told Business Insider the goal is to change how people interact with their computers. "Once ChatGPT can use computers and software faster than you or I can, it's going to change the way that you, by default, want to interact with your computer," he said.
Rival Anthropic was first to market with its own Computer Use feature in 2024. As of May, at least 600,000 organizations had tried Anthropic's Claude Cowork feature. OpenAI now says its tech has caught up, and it is betting that users will hand over more work to ChatGPT.
The training process for OpenAI's newest models included steps that improved their computer-use abilities. Zhou Yu, an AI researcher at Columbia University who worked on a benchmark test for computer-use agents, told Business Insider that the company likely included frame-by-frame screenshots in the earliest phase of model training. In a later step, developers provide the models with data showing the ideal inputs for a given output, such as filling out a tax form or 3D-modeling an object.
OpenAI has also upgraded how its models interact with a computer. Previously, ChatGPT would take a screenshot, analyze pixels, inject a command, and repeat. Now, when the tool has a website open, it can read the page's memory structure, accessibility information, and link connections directly. Weinstein said the mix of tools provides Codex with a way to test code, find bugs, and make improvements in a "full loop."
Cristian Medina Ruiz, a hobbyist coder in the Czech Republic, used Codex to rebuild the 2013 city-building game "SimCity" from its original source code. The Computer Use tool opened a window to verify that the game correctly simulated car movements and building images. Ruiz said he now leaves his code running and iterating when he is away from his computer.
Weinstein and James Sun, who works on browser capabilities at OpenAI, told Business Insider that the technology connects ChatGPT's agents to the "long tail" of websites built only for human hands. Users are automating data entry, compliance tasks, and calendar scheduling, they said.
Yu said there is a "gap" between the technology's current speed and what consumers want. "We definitely do see that computer-use agents could do well in limited domains, when we give it a lot of training data," Yu said. "When you want to generalize to any webpage, any application, any operating system, it's still very difficult."
Computer Use cannot churn through email responses, and scrolling through social media feeds remains laggy. Both Sun and Weinstein hope that as OpenAI's models improve, the experience will match coding, where it is often faster to let the agent write code than to do it manually.
When OpenAI CEO Sam Altman discussed an early version of Computer Use in 2025, he wrote on X that "the utility is significant" but "so are the potential risks." He recommended giving agents the minimum access required to complete a task.
Mark Beare, a general manager at the security firm Malwarebytes, told Business Insider the approach worries him. "This does still feel a little bit wild west-ish to me," he said, adding that customers may not have enough privacy protections. He said people should wall off passwords, sensitive data, and anything under a nondisclosure agreement, and avoid letting an agent run unsupervised on a long, complex task.
Sun said his team is working on "confirmation policy" – deciding when the tool should ask a user to permit its next step. For now, whenever an AI agent transmits or deletes data, it asks first. "You can make something 100% safe, but then it's just very, very difficult to use, and you can make something that is very easy to use but very dangerous," Sun said. "So we're really trying to figure out how to build something that's beneficial and also easy to use."
Drafted by a large language model from the source reporting linked above, then screened by automated publishing checks. It is not read by a journalist before publication. Some articles cite our Alpha Score. Verify prices and figures against the original source. Educational coverage, not personalized advice.