AI Powerhouse: Anthropic’s Claude Unravels Post-Quantum Encryption in Under 60 Hours
In a groundbreaking yet alarming display of artificial intelligence’s escalating capabilities, Anthropic revealed on July 28, 2026, that its yet-to-be-released Claude Mythos Preview model unearthed two novel attacks on cryptographic algorithms. The most concerning aspect is that one of these attacks was aimed at HAWK-256, a digital signature scheme that had withstood several rounds of expert human scrutiny within NIST’s post-quantum cryptography standardization process.
Operating largely independently within a multi-agent environment for approximately 60 hours—and at an approximate API compute cost of $100,000—the AI model drastically reduced the cost of stealing the smallest HAWK key from 2^64 operations to a mere 2^38. A human researcher offered only sporadic project-management guidance and was not a lattice-cryptography specialist. In a secondary discovery, Claude also devised a new technique it dubbed the Möbius Bridge, which accelerated an existing attack on a weakened version of AES between 200 and 800 times faster, surpassing a record held by cryptographers since 2013.
The repercussions were immediate: the HAWK team retracted the scheme from NIST’s standardization process the next day after verifying the attack approximately halves the block size necessary to recover an equivalent secret key. NIST promptly updated its third-round candidate page to mark HAWK as withdrawn. While experts point out that neither result disrupts production systems in common use today, the discoveries pose pressing questions about the pace at which AI could threaten the cryptographic bedrock of the modern internet.
Source: The Hacker News – Claude AI Just Cracked a Post-Quantum Test Scheme