The Pitfalls of "Security by Obscurity" And What They Mean for Transparent AI