The irony is spectacular: a coding assistant designed to automate development also automates the creation of security liabilities. And Anthropic’s response seems to be more automation. Perhaps the next AI agent can monitor the first AI agent while a third one explains to management why the database credentials are now public.