Anthropic's Mythos Model Cracked Cryptographic Algorithms — AI Found in 60 Hours What Experts Studied for 2 Years
Anthropic's Claude-based Mythos model discovered new attacks on HAWK post-quantum signatures and reduced AES, showing how AI could challenge the foundational assumptions of internet security.
AI Found What Humanity Has Been Missing for Years
Anthropic's new AI model, Claude Mythos Preview, has made a groundbreaking discovery in the world of cryptography. It found mathematical weaknesses in two major cryptographic algorithms — HAWK (a post-quantum digital signature scheme) and a reduced version of AES (Advanced Encryption Standard). What took human experts over two years, AI accomplished in just 60 hours.
The HAWK Attack: 60 Hours for $100,000
HAWK is one of the remaining candidates in NIST's competition for post-quantum signatures. These schemes are designed to remain secure even against future quantum computers. Human experts had reviewed HAWK for over two years, but Mythos found an improved attack that exploits a previously undetected symmetry in the mathematical lattice HAWK's security relies on. The model worked semi-autonomously in a multi-agent system — one agent initially dismissed the idea, but a second agent found a way to fully exploit it.
AES "Möbius Bridge" — 800x Better Attack
Mythos also found a new method for attacking a reduced version of AES-128 (7 out of 10 rounds). What makes this special: the AI developed a completely novel approach called "Möbius Bridge", which removes one of the guesses an attacker must make. The result improves on the best previously known attacks by a factor of 200 to 800.
The AI Initially Refused — "AES-128 r5/r6 is Just Genuinely Hard"
One of the most fascinating details is that Mythos initially refused to work on AES. The model wrote: "If you want a different outcome, the target has to change… AES-128 r5/r6 is just genuinely hard." After the researcher encouraged it to look for "genuinely novel ideas," Mythos began pursuing creative approaches. The human researcher had a background in theoretical computer science but wasn't an expert in lattice-based cryptography — his role was mostly project management.
What This Means for Internet Security
It's important to note: neither finding affects systems in use today. HAWK is only a candidate in the ongoing NIST standardization process, and the AES attack applies to a modified version using 7 of 10 rounds. Still, these results clearly show how AI models can challenge the core assumptions behind internet security.
Access to Mythos Remains Restricted
Anthropic has published two research papers on these findings and created CryptanalysisBench, a new benchmark for evaluating AI models' cryptographic analysis capabilities. However, access to Mythos Preview remains restricted. Anthropic continues to maintain a cautious approach to ensure this powerful tool doesn't fall into the wrong hands.