Anthropic's Claude Finds Weaknesses in HAWK and Reduced-Round AES

Anthropic, in collaboration with ETH Zurich, Tel Aviv University, and University of Haifa, released research showing its Claude Mythos Preview model achieved breakthroughs in cryptanalysis. The model ran for approximately 60 hours at an API cost of about $100k, discovering a stronger attack on the post-quantum digital signature scheme HAWK, significantly weakening its security, and a new attack on reduced-round AES that improves speed by 200-800x, including an improvement on existing attacks against 7-round AES-128. Anthropic published the related papers and AES chain-of-thought materials.

Confirmed

• Anthropic introduced CryptanalysisBench, a benchmark for LLM cryptanalysis capabilities, where frontier models can break 65%–86% of level-1 cipher schemes.
• Claude Mythos Preview found an attack on HAWK, which had undergone two years of expert review.
• The model found new attacks on reduced-round AES; standard AES-128 has 10 rounds, while the research targeted 7 rounds.
• Anthropic released the HAWK attack paper, AES attack paper, and model chain-of-thought materials.

Unconfirmed

• Claims that "AI broke post-quantum encryption" are exaggerated. Multiple experts (e.g., @muchcharles, @matthewdgreen) clarified: full AES remains secure; HAWK is not deployed and is harder to attack with larger parameters; the attack does not transfer to Falcon signatures. These findings do not directly affect current real-world systems.

Why it matters

• This marks AI's evolution from finding software implementation vulnerabilities to identifying mathematical flaws in cryptographic algorithms. Although currently theoretical and targeting simplified versions, it demonstrates the practical potential of large language models in complex cryptographic vulnerability discovery.

2026-07-29 ~ 2026-07-30 · 39 related posts

• Claude Exposes Novel Weaknesses in Cryptographic Algorithms — nordicinst · 2026-07-29
• [source] Anthropic says Claude found new weaknesses in HAWK and round-reduced AES — AnthropicAI · 2026-07-29
• [source] Anthropic’s new benchmark finds frontier LLMs can break 65%–86% of tier-1 ciphers — AnthropicAI · 2026-07-29
• [source] Anthropic posts the papers and chain-of-thought behind its crypto attacks — AnthropicAI · 2026-07-29
• Anthropic’s Claude Mythos reportedly found post-quantum crypto flaws humans missed — Polymarket · 2026-07-29
• Anthropic Shows Claude Can Discover Mathematical Flaws in Cryptographic Algorithms — moyix · 2026-07-29
• Mythos Preview found a new attack on HAWK after two years of review — Kyrannio · 2026-07-29
• Anthropic’s Claude Mythos AI reportedly finds new attacks on cryptographic algorithms — luisdans · 2026-07-29
• Anthropic's Mythos Preview Model Accelerates AES Attack Speed by 200-800x in One Week — tszzl · 2026-07-29
• Anthropic's Claude Halves Post-Quantum Security, Accelerates AES Attacks by 800x — petrusenko_max · 2026-07-29
• Anthropic says Claude Mythos Preview exposed structural weaknesses in HAWK cryptography — PsychologicalBox5208 · 2026-07-29
• Anthropic’s Claude found a missed structural weakness in HAWK after two years of review — ccerrato147 · 2026-07-29
• Claude AI Discovers Critical Flaw in Post-Quantum Security Candidate — Mazrael33 · 2026-07-29
• Curated Research on Post-Quantum HAWK and AES Cryptanalysis — imjustnewatai · 2026-07-29
• Anthropic researchers used Claude to find weaknesses in HAWK and weaker AES variants — muchcharles · 2026-07-29
• Anthropic says Claude helped uncover weaknesses in cryptographic algorithms — rickasaurus · 2026-07-29
• A one-line meme about Claude breaking AES lands after Anthropic’s crypto post — matthew_d_green · 2026-07-29
• Anthropic used Claude Mythos for 60 hours and found crypto flaws at a $100K cost — vista8 · 2026-07-29
• Anthropic says Claude found a faster attack on a reduced AES cipher, not real AES — nptacek · 2026-07-29
• Anthropic says Claude helped find cryptographic weaknesses, amid chain-of-thought debate — basedjensen · 2026-07-29
• Anthropic says Claude Mythos Preview helped find weaknesses in cryptographic algorithms — jedisct1 · 2026-07-29
• Claude Mythos Preview finds new cryptanalytic attacks on HAWK and AES — 新智元 · 2026-07-29
• Anthropic joke claims Claude Mythos found post-quantum crypto flaws — tanduyha · 2026-07-29
• Anthropic’s Claude shows strong cryptanalysis results on HAWK and AES — matthew_d_green · 2026-07-29
• Follow-up says HAWK is undeployed, non-transferable, and not just broken by AI — matthew_d_green · 2026-07-29
• Anthropic says Claude helped researchers find weaknesses in cryptographic algorithms — teortaxesTex · 2026-07-29
• Anthropic Hawk AES cryptanalysis thread draws strong praise — joshua_saxe · 2026-07-29
• Anthropic’s latest crypto attacks target HAWK and reduced-round AES, not deployed standards — matthew_d_green · 2026-07-29
• Cryptographer Analyzes Anthropic's New Cryptanalysis Results — matthew_d_green · 2026-07-30
• Anthropic Frontier Models Expose Vulnerabilities in Bank Encryption Algorithms — ChuckDBrooks · 2026-07-30
• Anthropic Research: Claude Discovers Weaknesses in Cryptographic Algorithms — AccBalanced · 2026-07-30
• AI Model Improves Attack on 7-Round AES, Sponsored by Anthropic with $100k Compute — jedisct1 · 2026-07-30

7 near-duplicate retellings: jedisct1 · ccerrato147 · fkasummer · inductionheads · matthew_d_green · vista8 · Assix0098