Claude can use your computer
What happened
Anthropic rolled out an update letting Claude natively use an opted‑in user’s computer to open documents and control apps — one of the first enterprise‑grade agent features with OS‑level integration announced March 24, 2026. Anthropic frames the capability as cautious and opt‑in, but it signals a shift toward agents that bridge cloud models with local workflows and raises new security tradeoffs. (cnbc.com)
Why it matters
Anthropic released the desktop-control research preview to Claude Pro and Max subscribers on macOS, exposing the capability through its Claude Code and Claude Cowork desktop apps. (engadget.com) Tasks can be dispatched from a phone into the desktop session using Anthropic’s Dispatch workflow, letting mobile prompts trigger the agent to click, type, and navigate desktop apps. (indianexpress.com) Anthropic tied the update to recent model upgrades—Sonnet 4.6 (announced Feb. 17, 2026) and Opus 4.6—highlighting a 1M‑token context beta in Sonnet to support longer, multi‑step UI interactions. (anthropic.com) The company says Sonnet’s computer‑use performance jumped on benchmark OSWorld from below 15% in late‑2024 to roughly 72.5% in its latest evaluations, a metric Anthropic published while describing its computer‑use work. (anthropic.com) Anthropic acquired Seattle startup Vercept on Feb. 25, 2026, folding Vercept’s desktop‑agent technology and team into Claude; the deal’s financial terms were not disclosed in coverage of the acquisition. (techcrunch.com) Anthropic’s announcement and system cards point developers to experiment with computer‑use via the Claude API and earlier Bedrock and Vertex AI integrations, and name Asana, Canva, Replit, DoorDash, and Cognition as early partners testing the feature. (anthropic.com) Anthropic’s public materials describe the capability as experimental and publish safety and performance evaluations for agentic workflows, while positioning Opus 4.6 as having a safety profile comparable to or better than other frontier models. (anthropic.com) News outlets place the rollout inside a broader industry sprint toward agentic desktop automation following viral demos like OpenClaw, noting Anthropic’s move comes amid intensified competition and talent deals in the agent space. (cnbc.com)
Key numbers
- Anthropic rolled out an update letting Claude natively use an opted‑in user’s computer to open documents and control apps — one of the first enterprise‑grade agent features with OS‑level integration announced March 24, 2026.
- (indianexpress.com) Anthropic tied the update to recent model upgrades—Sonnet 4.6 (announced Feb.
- 17, 2026) and Opus 4.6—highlighting a 1M‑token context beta in Sonnet to support longer, multi‑step UI interactions.
- (anthropic.com) The company says Sonnet’s computer‑use performance jumped on benchmark OSWorld from below 15% in late‑2024 to roughly 72.5% in its latest evaluations, a metric Anthropic published while describing its computer‑use work.
Quick answers
What happened in Claude can use your computer?
Anthropic rolled out an update letting Claude natively use an opted‑in user’s computer to open documents and control apps — one of the first enterprise‑grade agent features with OS‑level integration announced March 24, 2026. Anthropic frames the capability as cautious and opt‑in, but it signals a shift toward agents that bridge cloud models with local workflows and raises new security tradeoffs. (cnbc.com)
Why does Claude can use your computer matter?
Anthropic released the desktop-control research preview to Claude Pro and Max subscribers on macOS, exposing the capability through its Claude Code and Claude Cowork desktop apps. (engadget.com) Tasks can be dispatched from a phone into the desktop session using Anthropic’s Dispatch workflow, letting mobile prompts trigger the agent to click, type, and navigate desktop apps. (indianexpress.com) Anthropic tied the update to recent model upgrades—Sonnet 4.6 (announced Feb. 17, 2026) and Opus 4.6—highlighting a 1M‑token context beta in Sonnet to support longer, multi‑step UI interactions. (anthropic.com) The company says Sonnet’s computer‑use performance jumped on benchmark OSWorld from below 15% in late‑2024 to roughly 72.5% in its latest evaluations, a metric Anthropic published while describing its computer‑use work. (anthropic.com) Anthropic acquired Seattle startup Vercept on Feb. 25, 2026, folding Vercept’s desktop‑agent technology and team into Claude; the deal’s financial terms were not disclosed in coverage of the acquisition. (techcrunch.com) Anthropic’s announcement and system cards point developers to experiment with computer‑use via the Claude API and earlier Bedrock and Vertex AI integrations, and name Asana, Canva, Replit, DoorDash, and Cognition as early partners testing the feature. (anthropic.com) Anthropic’s public materials describe the capability as experimental and publish safety and performance evaluations for agentic workflows, while positioning Opus 4.6 as having a safety profile comparable to or better than other frontier models. (anthropic.com) News outlets place the rollout inside a broader industry sprint toward agentic desktop automation following viral demos like OpenClaw, noting Anthropic’s move comes amid intensified competition and talent deals in the agent space. (cnbc.com)
