Anthropic's Claude Mythos Cracks Two Encryption Systems in 60 Hours
Claude Mythos Preview cracked a post-quantum signature scheme and accelerated an AES attack by up to 800x, doing in days what human experts missed for years
- Claude Mythos Preview found a previously unknown attack on HAWK, a post-quantum signature candidate, cutting its effective key strength in half in just 60 hours.
- The model also improved attacks on 7-round AES by 200-800x using a novel technique it invented called the Möbius Bridge.
- Both results were achieved mostly autonomously, with minimal human prompting, at roughly $100,000 in API cost each.
- Anthropic released CryptanalysisBench, a 191-task benchmark across 6 cipher families to track LLM cryptanalysis capabilities over time.
- Neither attack breaks production systems: HAWK is undeployed, and the AES result targets a reduced 7-round variant, not the full cipher.
- Demonstration code for the HAWK attack is on GitHub; Anthropic also published Claude's raw chain-of-thought during the AES discovery.
Cryptography has always been a game of adversarial review: propose an algorithm, let the world attack it, and trust what survives. New research from Anthropic suggests that AI is about to dramatically accelerate that game. Using Claude Mythos Preview, Anthropic researchers have found meaningful weaknesses in two cryptographic systems -- one a post-quantum signature candidate still under government review, the other a reduced variant of the most battle-tested cipher in existence.
Why this domain, why now
Cryptographic algorithms are a fundamental building block of digital security. When you visit a webpage, your browser checks authenticity using a digital signature scheme, and traffic is encrypted using symmetric ciphers. Flaws in these widely used systems could put billions of users' data at risk. Until recently, finding those flaws required years of expert human effort. The question Anthropic set out to answer: could a frontier AI model compress that timeline?
When Anthropic launched Claude Mythos Preview, it was already able to autonomously find and exploit vulnerabilities in almost every piece of software pointed at it, including several major cryptographic libraries. Those vulnerabilities were due to incorrect implementation of algorithms -- errors in how programmers used the algorithms in their code. This new work goes a level deeper: finding mathematical flaws in the algorithms themselves.
Two targets, two breakthroughs
The first target was HAWK, a digital signature scheme designed to be secure against quantum computers. In 2022, NIST put out a call for additional cryptographic systems that would remain secure even against quantum computers, which could break most existing signature schemes in use today. HAWK is one of the third-round candidates under consideration. It had survived two years of expert human review.
Despite HAWK having survived two rounds of expert human review over a period of two years, Mythos was able to improve the best-known attack on it in just 60 hours of work -- effectively cutting its key strength in half.
The second target was a reduced-round variant of AES (Advanced Encryption Standard), the symmetric cipher that secures most of the internet's encrypted traffic. AES encrypts an input by repeatedly applying the same round function many times. AES-128 has 10 rounds; the attack works on a modified version with 7 out of the full 10 rounds. Academics regularly study round-reduced ciphers to gain insights into attack techniques that could generalize to the full cipher.
Mythos was able to develop an improved attack that extends a long line of research papers aiming to find the best attack on 7-round AES using a technique known as a meet-in-the-middle attack. At a high level, these attacks work by trading off time for space: by storing intermediate calculations and re-using them, it is possible to significantly reduce the runtime at the cost of constructing a large lookup table.