● Covered by 3 sources · 5 reportsMedium impact5 neutral
Anthropic's Claude Mythos AI Identifies Weaknesses in HAWK and Round-Reduced AES
Key points
- Claude Mythos Preview found a key-recovery attack against HAWK-256.
- HAWK, a post-quantum signature candidate, was withdrawn from NIST consideration.
- Mythos also found a 200- to 800-fold speedup for an attack on seven-round AES-128.
- Neither discovered attack affects currently deployed production systems.
- The research highlights AI's potential in cryptanalysis and identifying algorithm flaws.
AI Discovers Cryptographic Vulnerabilities
Researchers at Anthropic, using their Claude Mythos Preview AI model, have identified improved attack methods against cryptographic algorithms. The AI discovered a key-recovery attack against HAWK-256, a digital signature scheme designed for post-quantum security, and a significantly faster attack on seven-round AES-128, a widely used symmetric cipher.
HAWK Scheme Withdrawn from Standardization
The attack on HAWK exploits a previously unused symmetry in the lattice behind the signature scheme. Anthropic's released implementation of this attack has an expected end-to-end runtime of approximately three hours and 42 minutes on a 96-core server. Following Anthropic's announcement, the developer of HAWK withdrew the scheme from consideration in the National Institute of Standards and Technology (NIST) post-quantum standardization process. HAWK had survived two rounds of NIST testing and was in a third round when the flaw was uncovered.
Improved Attack on Round-Reduced AES
For AES, Claude Mythos Preview achieved a 200- to 800-fold speedup for an attack on seven-round AES-128. This result removes a 256-way guessing step from an existing meet-in-the-middle attack. The attack applies to seven of AES-128's ten rounds and still requires an impractical number of chosen plaintexts.
No Immediate Impact on Production Systems
Anthropic stated that neither the HAWK nor the AES result affects production systems currently in use. The public recovery code for HAWK only targets the smaller HAWK-256 parameter, and the AES attack does not apply to the full 10-round version of AES-128. No production software changes are required as a direct result of these findings.
Implications for Cryptographic Research
These findings represent significant research advancements, demonstrating AI's capability to identify mathematical flaws in cryptographic algorithms themselves, rather than just implementation errors. Anthropic published the findings alongside two technical papers and reproducibility artifacts, noting that Mythos Preview largely conducted the research, with human input for project direction and computing resources. This work suggests that large language models (LLMs) can aid in formalizing attack techniques and identifying errors in security proofs, serving as a new tool for cryptographic research.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub anthropics/cryptography-research-demo arXiv 2607.18538 CVE CVE-2026-454457.5 HIGH CVE CVE-2026-51949.1 CRITICAL CVE CVE-2026-505229.8 CRITICAL GitHub WordPress/WordPressHow outlets covered it
Anthropic's Claude Mythos LLM discovered new cryptanalytic attacks on HAWK and 7-round AES-128, demonstrating LLMs' potential in cryptanalysis without threatening current symmetric encryption standards. The work suggests LLMs can aid in formalizing attack techniques and identifying errors in security proofs. This indicates a new tool for cryptographic research, rather than a threat to existing secure algorithms.
The HAWK digital signature scheme, a candidate for a US quantum-resistant standard, was withdrawn from NIST consideration after Anthropic's Mythos AI security model identified a flaw. This development highlights the potential of AI in cryptanalysis and impacts the ongoing effort to establish post-quantum cryptographic standards.
Anthropic's unreleased Claude Mythos advanced AI model produced a new key recovery algorithm against HAWK, a proposed post-quantum signature scheme. This attack roughly halves HAWK's security bits, making the scheme less efficient and harder to justify for future standardization.
Anthropic's Claude Mythos Preview AI has discovered a key-recovery attack against HAWK-256 and a significantly faster attack on seven-round AES-128. These findings demonstrate AI's capability in cryptanalysis, though neither attack affects current production systems.
Anthropic's Claude Mythos Preview identified new attack methods against the HAWK digital signature scheme and round-reduced AES. These findings represent significant research advancements in cryptography, demonstrating AI's capability to find mathematical flaws in algorithms themselves, rather than just implementation errors.