Anthropic has launched a computer use feature for Claude Code, enabling it to take direct control of a desktop to complete tasks. The feature includes safeguards against prompt injection and blocks certain high-risk app categories by default, though Anthropic explicitly acknowledges these guardrails are imperfect and Claude may act outside them. The rollout positions Claude Code alongside a wave of similar personal computer agents from Perplexity, Manus, and Nvidia, with the broader trend tracing back to the viral spread of OpenClaw.
Products
Claude Code can now take over your computer to complete tasks
Anthropic launches computer use for Claude Code with acknowledged imperfect safeguards, joining competitors like Perplexity and Nvidia in the emerging desktop-agent market.
Wednesday, March 25, 2026 12:00 PM UTC2 MIN READSOURCE: Ars TechnicaBY sys://pipeline
Tags
products
/// RELATED
Safety6d ago
Bugs Rust won't catch
Canonical discovered 44 CVEs in Rust's uutils that bypassed Rust's entire safety model, proving the borrow checker can't prevent privilege-sensitive systems bugs like TOCTOU and symlink attacks—forcing Ubuntu 26.04 LTS to revert to GNU coreutils.
ResearchApr 22
DW-Bench: Benchmarking LLMs on Data Warehouse Graph Topology Reasoning
LLMs still struggle with structured database reasoning—DW-Bench is a new arXiv benchmark that measures how well they can navigate complex data warehouse schemas and topology.