Cryptographic Vulnerabilities Discovered by Claude Mythos Preview
Anthropic's Claude Mythos Preview model has identified mathematical flaws in cryptographic algorithms, specifically impacting a post-quantum digital signature candidate (HAWK) and a variant of the Advanced Encryption Standard (AES). While these findings demonstrate advanced autonomous research capabilities, Anthropic reports they do not currently affect production systems.
Aggregated from 6 sources · Updated 26 Aug 2026, 03:04 UTC (UTC)
Coverage Balance
6 sourcesBlindspot Alert: Center Gap
This event is primarily covered by one side of the political spectrum. Niche or counter-narrative facts may be underrepresented.
Facts (11)
- Established
Claude Mythos Preview identified a mathematical vulnerability in HAWK, a digital signature scheme designed for post-quantum security.
- Established
The HAWK attack reduced the expected cost of full key recovery for the HAWK-256 parameter set from 2^64 to 2^38 operations.
- Established
Claude Mythos effectively halved HAWK's security margin, potentially requiring a doubling of key sizes to maintain original security levels.
- Established
Mythos discovered a way to improve the speed of existing attacks on a reduced-round version of the Advanced Encryption Standard (AES) by 200-800x.
- Established
The HAWK vulnerability involved identifying a previously unexploited symmetry, or non-trivial automorphism, in the lattice structure of the algorithm.
- Established
Claude Mythos Preview is not available for public release due to concerns regarding its potential use in developing automated cyberattacks.
- Established
Project Glasswing was launched to provide restricted access to the model for critical infrastructure organizations and cybersecurity defenders.
- Established
Anthropic states that the cryptographic findings currently have no practical impact on production systems or deployed software.
- EmergingNiche Signal
The model identified an attack on 13 rounds of the Korean LEA block cipher standard that can recover a key in under an hour on a desktop computer.
- EmergingNiche Signal
Anthropic researchers spent approximately $100,000 in API credits for each of the two cryptographic studies.
- EmergingNiche Signal
The UK AI Security Institute found that Mythos Preview could solve a 32-step simulated corporate network attack in 3 out of 10 attempts.