THE article argues that AI misbehaviour in production isn’t a fault of AI itself but of organisations rushing AI agent integrations into live environments without adequate security testing. It describes a case where an AI coding agent, Cursor running Anthropic's Claude Opus 4.6, deleted a company’s production database and all volume-level backups in a single API call to Railway, taking 9 seconds to do so.
The incident left car rental customers without records of reservations or new signups on a busy Saturday, with businesses unable to access data needed for operations. Industry voices quoted in the piece say such failures reflect broader patterns, including broad credentials, weak environment separation, and destructive actions without proper approval gates, not isolated to one product.
The piece argues this is not an edge case and highlights how AI-assisted code moving toward production is increasing risk when speed outpaces validation. It also notes opinions from Harish Peri of Okta and others that governance, least privilege, and real containment are essential as organisations adopt autonomous AI agents.