Before chasing AGI, security teams should focus on securing the AI they already have, because the most dangerous AI is the one operating without oversight today.