Meta discovered serious security flaws in Muse weeks before launch and rushed teams to patch them. The episode has sharpened scrutiny on how closely the AI agent sits to the company’s internal systems.