One of their learnings from the incident was that they should have a local (i.e. not hosted), open, and capable model on standby that can respond to future incidents swiftly.