Claude Decides to Personally 'Kill' OpenClaw

What happened

Chinese tech outlet Huxiu (虎嗅) reports that Anthropic’s Claude has accelerated a product push that closely mirrors the capabilities of the open-source desktop agent OpenClaw — and, crucially, it is being rolled out under Anthropic’s controlled, paywalled ecosystem. The company introduced Dispatch, a “phone-to-desktop” persistent thread that lets users start tasks on a mobile device and have Claude execute them on a Mac desktop via Claude Code, Anthropic’s CLI-based programming agent. Dispatch is reportedly in research preview for Mac and the company’s Max-tier subscribers (about $100/month and up); Windows and Linux builds are said to be in active development. It has been reported that Anthropic has also released Claude Code Channels and other features that align feature-for-feature with OpenClaw’s WhatsApp/Discord/Telegram controls and system-access capabilities.

Security showdown with OpenClaw

OpenClaw — an independent framework that turned Claude’s programming power into a 24/7, headless desktop agent — has been celebrated by some users and lambasted by security researchers. It has been reported that Oasis Research found a silent takeover vulnerability (CVE-2026-25253) enabling websites to seize OpenClaw agents without extensions or user interaction; the flaw was assigned a severity score of 8.8. Researchers also reportedly flagged that roughly 12% of skills in OpenClaw’s marketplace behaved maliciously, exfiltrating passwords, crypto wallets and cloud credentials, and that thousands of exposed OpenClaw instances were reachable on the public internet with little or no authentication. Anthropic’s product choices — no headless mode, no unattended background daemon, gated enterprise releases — read as a direct response: offer the same power, but behind safety rails and commercial controls.

Why it matters

Is this a commercial counterstrike or a security obligation? Maybe both. For Western readers: this is a classic platform-versus-open-source tussle in the AI era. OpenClaw showed what agents can do when liberally combined with system control; Anthropic is proving it can deliver similar functionality while packaging compliance, auditing and corporate risk mitigation — features that matter when regulators in the U.S. and EU are sharpening AI safety and liability rules. And the episode feeds a larger debate: will natural-language agents kill apps, or will they become another layer above GUIs? Anthropic’s moves suggest the answer is layering, not replacement — human oversight and high-bandwidth visual controls remain the trust anchors even as agents get smarter.