Loading story…
Anthropic disclosed that its Claude Mythos Preview model helped uncover two new cryptanalytic results: an end-to-end key-recovery attack against the post-quantum signature target HAWK-256 and a substantially faster attack on 7-round AES-128. In the HAWK work, researchers said the model identified a previously unused symmetry in the scheme’s lattice structure, cutting the estimated cost of breaking HAWK-256 from about 2^64 to 2^38 operations and making recovery practical for that challenge parameter, while HAWK-512 and HAWK-1024 remain impractical to attack.
Anthropic said the AES result improved an existing meet-in-the-middle attack by removing a 256-way guessing step through a new fingerprinting technique described as a Möbius Bridge, yielding a roughly 200- to 800-fold speedup against reduced-round AES-128. The company emphasized that full 10-round AES-128 remains secure and that neither finding requires changes to production systems, but said the results show frontier AI can materially accelerate cryptanalysis; Anthropic coordinated disclosure of the HAWK finding with the scheme’s authors and NIST, and released technical papers, code, reproducibility artifacts, and a new CryptanalysisBench benchmark.
Track how attackers are adapting to this technology.
4 events from the most recent confirmed update back to the earliest known activity.
Anthropic released technical papers, code, and reproducibility materials for the cryptanalysis work, and launched CryptanalysisBench with academic partners to benchmark language models on cryptanalysis tasks. These releases accompanied the public disclosure of the results.
Anthropic reported that Claude Mythos Preview helped derive an improved key-recovery attack against the HAWK-256 post-quantum signature challenge and a much faster attack on reduced-round 7-round AES-128. The company said the findings do not threaten deployed production systems or full 10-round AES-128.
After confirming Anthropic's findings, the authors of HAWK withdrew the post-quantum signature scheme from NIST's additional standardization program. They said straightforward mitigations such as doubling parameters would make HAWK uncompetitive.
Anthropic said it disclosed the HAWK weakness to the scheme's authors and coordinated the finding through NIST. This reflects the formal responsible-disclosure step tied to the newly identified lattice-symmetry weakness.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
17 references tracked. Mallory keeps watching after this page renders.
livescience.com
Open sourcexakep.ru
Open sourcezdnet.fr
Open sourcearstechnica.com
Open sourceanthropic.com
Open sourceanthropic.com
Open sourcenist.gov
Open sourceanthropic.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Anthropic said its `Mythos` AI security model found a weakness in the post-quantum digital signature candidate **HAWK**, prompting the algorithm’s withdrawal from NIST consideration after the system reportedly identified a new way to locate automorphisms that cut HAWK’s effective security roughly in half. The same research also improved a meet-in-the-middle attack on a deliberately weakened seven-round AES challenge, reducing the known-plaintext requirement from about `2^105` to `2^89`, although both results were described as impractical against widely deployed production cryptography and focused on challenge systems or immature candidates rather than live enterprise deployments. The developments come amid broader concern that AI is changing how cryptanalytic claims are validated, because multiple researchers may rely on the same models and inherit common failure modes while appearing independent. Recent examples cited in the debate include near-simultaneous July papers on unclonable encryption that both credited OpenAI’s `GPT-5.6 Sol Ultra` with core ideas, and AI-assisted reviews of a claimed polynomial-time algorithm for the Dihedral Coset Problem that disclosed model involvement unevenly; researchers are calling for clearer provenance, explicit verification-status reporting, and continued emphasis on **crypto-agility** rather than abrupt changes to post-quantum deployment plans based on early AI-assisted findings.
Aug 15, 2026
Researchers reported that **HAWK**, a lattice-based digital signature candidate in NIST's additional post-quantum signature process, can be weakened when an attacker knows a nontrivial automorphism of the underlying integer lattice. In the paper *"HAWK: Having Automorphisms Weakens Key"*, the authors show that for cyclotomic rings whose degree is a power of two, the search rank-2 module Lattice Isomorphism Problem used by HAWK can be reduced to an instance with at most half the rank, accelerating key-recovery attacks by at least a quadratic factor and effectively cutting the scheme's security margin by roughly half in bits. The work does **not** claim a full practical break of HAWK, but it identifies a new attack avenue tied to the automorphism group of the lattice structure. The paper also links its findings to earlier ASIACRYPT 2024 research by Luo et al., which identified a symplectic automorphism that broke HAWK's original `omSVP` assumption; the HAWK team subsequently revised that definition in its second-round NIST additional-signatures submission. HAWK remains listed among NIST's additional digital signature candidates, making the result relevant to organizations tracking the resilience of post-quantum signature schemes under evaluation.
Jul 29, 2026
Researchers introduced **CryptanalysisBench**, a 191-task benchmark spanning six families of cryptographic primitives, to measure whether large language models can discover practical cryptanalytic attacks against historical and production schemes. In reported results, five frontier models broke **65% to 86%** of Tier 1 schemes, found attacks against **6 to 12** Tier 2 schemes at full strength, and solved **24 to 61** scaled-down Tier 2 variants, suggesting that AI capability in cryptanalysis is advancing beyond toy examples. The work also reported **novel cryptanalytic findings**, including a **key-recovery attack** on the **SpoC AEAD** scheme and identification of an error in **KINDI**’s published **CCA-security proof**. Commentary on the research said Anthropic used the benchmark to test **Mythos Preview**, which reportedly uncovered new vulnerabilities in **Hawk** and **reduced-round AES**, reinforcing concerns that AI-assisted cryptanalysis could become a meaningful security factor and a useful stress test for candidate cryptographic schemes before deployment.
Jul 29, 2026