Anthropic's LLM Discovers New Attack Paths for AES and HAWK
matthew_d_green · x · 2026-08-01
Cryptographer JP Aumasson reviewed Anthropic's recent breakthroughs in cryptanalysis using Large Language Models (Claude Mythos).
- HAWK Attack: Discovered a key-recovery attack against HAWK, a NIST post-quantum signature candidate, reducing the estimated security level of HAWK-512 from 128 bits to at most 108 bits (speculatively 81 bits).
- AES Attack: Found an improved key-recovery attack on 7-round AES-128 (the full version has 10 rounds). Although not a practical threat, it is academically noteworthy as new cryptanalytic results for AES are rare.
The author noted that while LLMs cannot currently break symmetric encryption, they show great potential in formalizing attack techniques, verifying complexity estimates, and finding errors in security proofs.
More from Research
- It's not gradient descent: burny_tech corrects the diagnosis to RLHF mode collapse — burny_tech · 2026-10-16
- A beginner-friendly introduction to causal inference kicks off, Part 1 — Nasereliver · 2026-10-16
- FBI method trains for Pass@K with one sample per update at 12.3% of the cost — EmtiyazKhan · 2026-10-16
- Berkeley researchers sample humanoid robot policies directly from weight space with diffusion — berkeley_ai · 2026-10-16
- OneWM: unified causal training shows world and action modeling improve each other — IAAI233 · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16