Marcel Velica discusses the risks of granting AI agents access to business systems, highlighting security concerns around malicious approval and unauthorized system access.