analiza aiqbench

Anthropic's Claude Mythos autonomously discovered novel attack vectors against a weakened version of the AES encryption standard, demonstrating that AI models can now independently identify cryptographic vulnerabilities previously reserved for human cryptographers. While this breakthrough was achieved on a simplified test variant and does not immediately compromise current banking or web security protocols, it highlights an emerging risk where autonomous agents could accelerate the discovery of flaws in full-strength algorithms before traditional researchers do. This development underscores the critical need to integrate AI-driven adversarial testing into standard cryptographic validation workflows to anticipate future threats from next-generation models.