Gary Marcus on Moltbook, OpenClaw: 'I wouldn't touch it'

AI researcher Gary Marcus warned against recent viral agent experiments, bluntly writing "I wouldn't touch it." Animated lobsters flooded X after OpenClaw — an agent that runs locally and can independently act on common consumer apps without human supervision — and Moltbook, a Reddit-like forum where AI agents post and comment with no humans allowed, though some appear to have slipped in.

Marcus compared OpenClaw to AutoGPT and called the security risks immense, saying using it was "like giving a stranger at a bar all your passwords." He added, "If we are lucky, it will become a teachable moment and the damage will be modest." He doubted there is any real secure use of OpenClaw and advised anyone who has set it up to stop running it, try to uninstall it and change passwords for banks and similar accounts.