AI models have identified mathematical vulnerabilities in HAWK and Classic McEliece, two post-quantum cryptographic schemes. Co-founder of Project Eleven, Conor Deegan, highlighted these findings on October 19, 2026, noting that the flaws were discovered without the use of quantum computers.
In July 2026, the Claude Mythos AI model by Anthropic detected a symmetry-based key-recovery attack on HAWK. This attack reduced the effective security of HAWK-512 from 2^150 to 2^108, leading to the scheme's withdrawal from the NIST standardization process.
Classic McEliece also faces scrutiny, with recent analyses estimating key-recovery complexity as low as 2^94 operations. Consequently, the German Federal Office for Information Security (BSI) advised against its deployment in early October 2026.
While no production systems using standardized algorithms have been compromised, these events underscore the role of AI in accelerating the discovery of cryptographic weaknesses. Project Eleven, founded in 2024, recently launched Strongpoint, a custody platform utilizing hybrid signing and modular algorithm integration to mitigate such risks.
