Loading story…
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.
Track how attackers are adapting to this technology.
1 event from the most recent confirmed update back to the earliest known activity.
The paper "CryptanalysisBench: Can LLMs do Cryptanalysis?" is published, introducing a 191-task benchmark for evaluating whether large language models can discover practical attacks against cryptographic schemes. The paper reports that frontier models broke many Tier 1 schemes, found attacks on some Tier 2 schemes, and produced novel results including a key-recovery attack on SpoC AEAD and identification of an error in KINDI's published CCA-security proof.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
2 references tracked. Mallory keeps watching after this page renders.
Map 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 published **Exploit Evals**, a framework for measuring whether advanced AI systems can carry out offensive cybersecurity tasks, as attention shifts from simple vulnerability identification to practical exploitation capability. The work highlights testing focused on realistic exploit-development steps and reflects growing concern that frontier models may increasingly assist with planning, chaining actions, and executing multi-stage attacks against software flaws. Separate benchmark results presented by Bugcrowd in **ExploitBench** found Anthropic’s **Claude Mythos** outperforming OpenAI’s **GPT-5.5** on exploitation tasks involving Google Chrome **V8** vulnerabilities. According to the reported results, Mythos averaged **9.90/16** and reached the highest exploitation tier on **21 of 41** vulnerabilities, while GPT-5.5 averaged **5.51/16** and reached that tier in **two** cases; researchers and industry experts said the findings suggest AI models are becoming more effective in offensive workflows and increase pressure on defenders to accelerate AI-assisted remediation, prioritization, and rapid patching.
Aug 11, 2026
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.
Aug 7, 2026
Security researchers and AI vendors reported rapid progress in applying frontier models to offensive and defensive cybersecurity work, including the discovery of **`CVE-2025-37899`**, a remote zero-day in the Linux kernel’s SMB implementation. A published account described using an advanced model to help identify the flaw, while OpenAI's **Daybreak** and Anthropic's **Project Glasswing** highlighted broader efforts to build AI systems for cyber operations, vulnerability research, and security workflows. SentinelLABS added evidence that these systems are moving beyond isolated tasks into sustained investigations, describing an eight-stage benchmark based on reverse engineering **fast16**, a 2005 Windows sabotage toolkit tied to interference with high-precision solvers used in nuclear-weapons modeling. In that evaluation, **GPT-5.6 Sol** was reported as the only publicly available model tested that completed the full long-horizon malware-analysis workflow, while other models showed useful point-in-time reasoning but failed to maintain consistency as new evidence forced earlier conclusions to be revised.
Jul 29, 2026