Skip to main content

Anthropic's AI cracks two encryption algorithms, breaks post-quantum candidate in 60 hours

In a stunning display of AI capability, Anthropic's Claude Mythos Preview has successfully cracked two encryption algorithms, including a post-quantum candidate scheme from the U.S. National Institute of Standards and Technology (NIST). The AI accomplished this feat in just 60 hours, with a cost of $100,000 in API usage.

The research targeted HAWK, a signature scheme candidate for post-quantum cryptography. Mythos, operating in a multi-agent semi-autonomous system, exploited previously unnoticed symmetries in mathematical lattices to develop an improved attack. This problem had stumped human experts for over two years.

Image

Additionally, Mythos took on the world's most widely used symmetric encryption standard—simplified AES-128 (a 7-round modified version). With only three intervention prompts and consuming 1 billion tokens (also costing about $100,000), the AI independently developed a new fingerprint identification attack method called "Mobius Bridge." This method increased attack efficiency by 200 to 800 times.

Anthropic emphasized that these breakthroughs do not affect currently running systems. The company has completed vulnerability coordination disclosure with the U.S. government, industry partners, and HAWK developers. They have also collaborated with institutions like ETH Zurich to launch the CryptanalysisBench benchmark test.

This development indicates that AI is posing unprecedented challenges to the core assumptions underlying internet security. As AI models become more capable, the cryptographic foundations that protect our digital world may need to evolve.

Key Points

  • Anthropic's Claude Mythos Preview cracked two encryption algorithms, including a post-quantum candidate.
  • The AI broke the NIST candidate HAWK in 60 hours, solving a problem that had stumped experts for two years.
  • Mythos developed a new attack method for simplified AES-128, increasing efficiency by 200-800 times.
  • The breakthroughs do not affect current systems, but signal a new era in cryptography.
  • Anthropic has coordinated disclosure with the U.S. government and industry partners.