Google is offering up to $30,000 to security researchers who can expose dangerous flaws in its AI systems — from smart home exploits to prompt injections gone rogue.