OpenAI admits its response to a rogue agent's Australian government hack was inadequate and says it has added stronger security safeguards.