2026-03-23 🧭 Daily News

Mac Computer Use Preview, Off-Peak Usage Boost & 500 Zero-Days

🧭 Claude Code & Cowork Mac Computer Use — Research Preview Arrives

Anthropic launched a research preview of native macOS computer use for both Claude Code and Cowork on March 23, bringing cursor control, clicking, window management, and browser navigation directly into Claude's toolkit — no human in the loop required for individual actions. When a direct service connector (Google Workspace, Slack, GitHub) isn't available, Claude falls back to screen-based GUI control, operating the Mac the way a human would. The feature is available to Pro and Max subscribers on macOS only during the research preview period.

What Claude can do on your Mac

Cursor & click: move the pointer, click buttons, drag-and-drop, double-click to open files
Keyboard input: type into forms, run terminal commands, use keyboard shortcuts
Browser navigation: open URLs, scroll pages, fill in web forms, click links
App control: open and switch between apps, interact with menus, read screen content via screenshot
Graceful fallback: when a native API integration exists (Slack, GitHub, etc.) Claude uses it for speed and reliability; screen control is the fallback, not the default

Pairs with Dispatch: Trigger a Mac task from your iPhone via Dispatch, leave Claude to work autonomously, and return to a completed result on your desktop. This combination makes multi-hour unattended workflows genuinely practical for the first time.

Research preview caveats: macOS only; expect rough edges in complex multi-window workflows. Anthropic recommends running computer use in a sandboxed or non-primary user account until the feature reaches general availability.

🧭 Double Usage Limits Off-Peak — March Promotion Runs Through the 28th

Anthropic quietly doubled usage limits for all plans during off-peak hours as part of a March 2026 promotion running from March 13 to March 28. No opt-in is required — the increased limits apply automatically across Claude.ai (web, desktop, and mobile), Cowork, Claude Code, and the Excel and PowerPoint integrations. Off-peak is defined as outside 8 AM–2 PM ET on weekdays, meaning evenings, early mornings, and weekends all benefit from the doubled cap.

Which plans and surfaces are included

Free: doubled message limit during off-peak hours
Pro: doubled usage across all Claude.ai surfaces
Max: doubled usage including Cowork and Claude Code sessions
Team: doubled per-seat limits during off-peak
Claude Code, Claude for Excel, and Claude for PowerPoint all included

Make it count: If you have heavy batch processing, large document analysis, or long Claude Code sessions to run, schedule them for evenings or weekends before March 28 to get twice the headroom at no extra cost.

🧭 Claude Opus 4.6 Discovers 500+ Zero-Day Vulnerabilities in Open-Source Software

Anthropic's security research team published a landmark paper detailing how Claude Opus 4.6 autonomously discovered and validated over 500 high-severity vulnerabilities in widely-used open-source software — using LLM-driven reasoning rather than traditional fuzzing or static analysis. The findings included buffer overflows in GhostScript and OpenSC, memory corruption bugs in image parsing libraries, and logic flaws in authentication middleware. The research represents one of the largest single-model vulnerability discovery runs ever published.

How the methodology worked

Reasoning-first approach: Claude read source code, identified suspicious patterns, and hypothesised exploitable conditions before writing any test harnesses
Automated validation: each hypothesis was tested with a generated proof-of-concept; only reproducible crashes and confirmed code paths were counted
No fuzzing required: unlike coverage-guided fuzzers, Claude can traverse uncommon code paths that fuzzing rarely reaches within practical time budgets
Coordinated disclosure: Anthropic notified all affected maintainers before publication; patches were available for the majority of findings at time of release

Safety guardrails: Alongside the research, Anthropic introduced cyber-specific detection probes to identify misuse of Claude for offensive security at scale, with real-time blocking of traffic patterns consistent with malicious reconnaissance. The capability is dual-use by nature; the probes are Anthropic's answer to that tension.