Researchers at Anthropic have reported that their advanced artificial intelligence model, Claude Mythos Preview, autonomously discovered novel methods to attack weakened versions of two significant cryptographic algorithms, demonstrating a potential new frontier in automated security analysis.