An Anthropic Claude AI Model Finds Flaws in Tough-to-Crack Encryption Algorithms