Zhou Hongyi warns desktop AI “OpenClaw” can hallucinate and pose data‑security risks, promises one‑click installer

Topline: a cybersecurity alarm from a veteran

Zhou Hongyi (周鸿祎), founder and chairman of Qihoo 360 (奇虎360), used his personal account to warn that the current wave of desktop AI agents brings real security risks. He said some implementations “hallucinate” and warned they could — in extreme cases — delete files on a user’s C: drive. It has been reported that community tests and early adopters have already flagged instability and data‑safety issues in similar local AI projects.

What's new — innovation meets growing pains

Zhou praised OpenClaw for making the abstract idea of an AI agent tangible — netizens have nicknamed it “养龙虾” (raising a lobster) because it can be hosted and “fed” on a user’s PC — but he also listed three pressing shortcomings: inadequate security, high installation barriers for ordinary users, and a skillset skewed toward high‑end scenarios rather than everyday tasks. He argued that most in‑house large models remain at the “chatbot” stage and that enterprise AI deployment will require people who understand both technology and business.

Bigger picture: why this matters beyond China

Zhou said his team will roll out a simplified one‑click installer to lower the barrier and push toward a future where “everyone has a personal AI assistant.” That push comes as China doubles down on local AI ecosystems amid U.S. scrutiny and export controls on advanced chips — a geopolitical backdrop that accelerates interest in on‑device and domestically hosted models. Can faster adoption happen without a costly security failure? Security veterans like Zhou are signaling that the answer still depends on closing significant technical and operational gaps.