Claude can now control your computer to carry out tasks on macOS

Anthropic’s Claude can now take over your screen and perform tasks for you, a move that puts the company in direct competition with a popular but harder-to-use open-source tool.

On Monday, March 23, 2026, the AI company announced that its Claude assistant could now click, type, access apps, and browse the web just like a human.

The feature lets Claude perform tasks such as finding and sending files stored on a hard drive, filling in spreadsheets, and navigating websites on a user’s behalf.

The announcement comes as tech companies scramble to build so-called “agents,” AI systems that can carry out tasks on their own without someone guiding each step. Much of the recent buzz around this idea started with OpenClaw, a free, open-source project built by Peter Steinberger.

OpenClaw allows people to give instructions to an AI through messaging apps like WhatsApp or Telegram. It gained popularity quickly, but setting it up is not simple. Users need to enter API keys, work through a terminal, and have some technical knowledge.

Anthropic is targeting users who found all of that too complicated.

The company says Claude works out of the box, with built-in connections to tools like Google Drive, Gmail, and Slack.

There is also a new feature called Dispatch, part of Anthropic’s Claude Cowork product, that lets someone assign a task from their phone, step away, and come back to find the job done on their desktop.

“Assign a task from your phone, turn your attention to something else, and come back to finished work on your computer,” the company wrote in a post on X.

Safety concerns take center stage

Where Anthropic is drawing the clearest line against OpenClaw is on the question of safety.

OpenClaw asks for broad access to a user’s entire system, which has made some people nervous about what an AI agent might do if something goes wrong, whether it deletes the wrong files, gets manipulated, or is used in an attack.

Anthropic says Claude takes a different approach.

Turning the feature on requires a single click. The arrangement requires users to authorize each app separately rather than granting them whole system access.

Additionally, the company has built-in defenses against prompt injection, a type of attack in which malicious actors attempt to trick an AI into executing dangerous instructions concealed within the text it consumes.

That said, Anthropic was upfront about the fact that this technology is not ready for everything.

The company told users not to let Claude access apps that hold sensitive data for now, and said the assistant will always ask before opening a new application. Business customers using Claude Code can also set rules about which outside tools are allowed.

Markets react as the race heats up

The ripple effects from this wave of AI agents have already been felt in financial markets.

Earlier this year, news that Anthropic’s tools could take over tasks handled by software companies sent Indian IT stocks lower.

Meanwhile, in China, where neither Anthropic nor OpenAI operates officially, OpenClaw spread widely. People there bought up Mac Mini computers to run the software themselves, and supplies reportedly sold out.

The Chinese government has since told state-owned companies to stop running OpenClaw on office machines.

While Claude sacrifices some raw flexibility for smart guardrails and plug-and-play convenience, OpenClaw offers power users unlimited freedom at the expense of safety and simplicity, making it the far safer and more accessible option for the majority of people.

OpenClaw’s creator, Peter Steinberger, was later hired by OpenAI. Microsoft and Google are also building their own versions of this kind of technology.

For now, Anthropic’s computer-use feature is available as a research preview on macOS for users on the Claude Pro and Claude Max plans.

Get seen where it counts. Advertise in Cryptopolitan Research and reach crypto’s sharpest investors and builders.