OpenClaw experts say the viral AI agent is overhyped, Moltbook posts were likely human-written

Security researchers found that Moltbook's Supabase credentials were unsecured, allowing anyone to impersonate AI agents. The supposedly autonomous AI expressions of angst were likely written by humans or prompted with human guidance. Permiso Security CTO Ian Ahl confirmed the vulnerability. The finding deflates excitement from figures like Andrej Karpathy, who called Moltbook activity "the most incredible sci-fi takeoff-adjacent thing." Multiple AI experts told TechCrunch that OpenClaw's capabilities are not as novel as the hype suggests.

View full digest for February 17, 2026