Claude Begins Challenge to Global Encryption System, but the Internet Remains Secure

According to Dynamic Watch Beating Monitor, Anthropic enables Claude Mythos Preview to autonomously explore research papers, formulate hypotheses, write code, and run experiments to identify mathematical weaknesses in cryptographic algorithms. It introduces two new attacks: one targeting the yet-to-be-deployed post-quantum signature scheme HAWK and another targeting a simplified version of AES. The former took about 60 hours to complete, while the core idea of the latter was mainly proposed by the model.

HAWK can be understood as digitally stamping a file, designed to withstand future quantum computers. Claude breached the HAWK-256, which was specifically used for testing, and lowered the security estimation for larger parameters. The value lies in proactively identifying issues in candidate algorithms, not in breaking into systems already in operation.

AES is the encryption standard widely used by websites, banks, and others, but the full AES has not been compromised. The experiments only targeted a 7-round version with 3 rounds of protection removed, requiring approximately 2^105 sets of specialized inputs. The so-called 200 to 800 times speedup only accelerates a theoretically unfeasible attack.

Claude is now able to engage in cryptographic research but is not yet capable of independent research. It managed to find key insights into AES within a few days, while the Anthropic research team spent hundreds of hours verifying it over nearly a month. At present, Claude can expedite research but cannot replace experts, let alone directly breach real-world systems.