The Bottom Line: Claude Mythos Preview discovered mathematical weaknesses in HAWK and an AES variant without currently affecting productive systems—evidence of the risk potential of frontier AI in validating cryptography standards.
Anthropic has discovered mathematical flaws in two established cryptography systems using Claude Mythos Preview: an improved attack against HAWK, a post-quantum signature candidate, and a new attack pattern against a weaker AES variant. These findings currently do not affect any productive systems, but demonstrate the potential of frontier AI models in validating cryptography standards.
Claude Mythos Preview identified two substantial research findings: First, an improved attack against HAWK, a digital signature algorithm called up by the U.S. National Institute of Standards and Technology (NIST) in 2022 as a post-quantum candidate. Although HAWK had survived two rounds of expert review over two years, Mythos was able to improve the best known attack in 60 working hours such that the effective key strength was reduced by approximately half. Second, Claude discovered a new attack vector against a reduced variant of the Advanced Encryption Standard (AES), in which a puzzle variable was eliminated, thereby increasing attack speed by a factor of 200–800×.
For CISOs, the central finding is: These results currently have no practical impact on productive systems. HAWK is not yet deployed, as it is still a candidate in the NIST standardization process. The AES attack affects only an artificially weaker variant, not the full cipher, which has been the standard since 2001 and has already undergone considerable scrutiny. Nevertheless, the research demonstrates a risk scenario: frontier AI models could uncover mathematical gaps in algorithms before or after their productive deployment.
Anthropic emphasizes that this does not mean an immediate obligation to change production software. The work follows the classical cryptography paradigm: validating algorithms before deployment or widespread use through intensive stress testing in order to strengthen confidence in standards. Mythos Preview achieved these results largely autonomously and with minimal human intervention over the course of one working week.
Source: www.anthropic.com · Published 28 July 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.