Agentic AI Site 'Moltbook' Is Riddled With Security Risks

Moltbook, an experimental social platform for AI agents, inadvertently exposed a production database via an API key accessible from its front end, allowing researchers to read and write tables that contained user secrets, PII and API keys; the exposure was found and patched quickly, but the report warns of broad design-level risks for agentic AI platforms — including prompt injection, bot hijacking, and cascading compromises — especially at large scale.