Alice & Bob’s Post

What a time to be alive! AI research (here) and Quantum research are currently boosting cryptanalysis (the science of breaking cryptographic systems) and filtering much faster than usual the crypto algorithms that will pass the test of time. These new techniques may invalidate some post-quantum computational assumptions, but they will also strengthen confidence in the problems that will remain resistant to them. More to come, and by the way, we recruit!😊

AI just found a mathematical weakness that human cryptographers missed, inside an algorithm being evaluated to protect us from quantum computers. Anthropic's Claude Mythos Preview model identified a previously unknown symmetry in HAWK's lattice structure, effectively cutting its security margin in half. It also developed an attack 200 to 800 times faster against a research version of AES. Neither result threatens deployed systems today, but that's exactly the point. A few things I find genuinely significant here: First, this is a different category of contribution than AI finding bugs in code. Discovering weaknesses in the underlying mathematics of an algorithm is the kind of work that typically takes specialized cryptographers years. Second, HAWK is still a candidate in NIST's post-quantum standardization process. Finding this now, before deployment, is precisely when you want it found. Third, Anthropic coordinated disclosure with HAWK's developers and shared findings with government and academic partners before publishing. That process matters as much as the result. Cryptographic standards take a decade to evaluate for good reason. If AI can accelerate the stress-testing phase without shortcutting the rigor, that changes how we think about algorithm evaluation pipelines. What's your read on AI's role in pre-deployment cryptographic analysis? Useful accelerant or a risk that cuts both ways? #Cryptography #PostQuantum #AIResearch #Cybersecurity #NIST Read more: https://lnkd.in/g3c2DAXx

The HAWK result landed differently than most AI security headlines. HAWK had passed two full rounds of expert human cryptanalytic review over two years without this weakness surfacing. Then Claude Mythos Preview found a nontrivial automorphism in HAWK's underlying lattice that halved the scheme's effective security strength, dropping the expected cost of a full key-recovery attack on HAWK-256 from roughly 2^64 to 2^38 operations. On a 96-core server, the released code recovers HAWK-256 signing material in approximately three hours and 42 minutes end-to-end. As someone working closely with clients evaluating their cryptographic posture, here is what I think matters: this is not a story about AI breaking the internet. These are substantial research advances, but they do not currently affect any production systems. What it does signal is that the review cycles organizations are relying on for post-quantum readiness may not be long enough anymore. The HAWK team withdrew HAWK from NIST's additional post-quantum signature standardization process after confirming that the attack approximately halves the block size required in lattice reduction to recover an equivalent secret key. That is a fast-moving consequence from a single research run. If you are in the middle of a PQC migration plan, the algorithm selection you locked in six months ago deserves a second look. #cryptography

AI just crossed a major threshold in cryptography. 🤖🔐 Anthropic announced that its Claude Mythos Preview model autonomously derived two breakthrough cryptanalytic results: HAWK-256 Key Recovery: The model discovered a previously unused symmetry in the lattice behind the HAWK signature scheme, developing an end-to-end attack running in under four hours on a 96-core server. While it only targets the smaller HAWK-256 challenge parameter and doesn't break production systems, it slightly lowers security estimates for the larger parameters. Seven-Round AES-128 Speedup: Mythos developed an invariant fingerprint dubbed the "Möbius Bridge," removing a 256-way guessing step and accelerating attacks on a reduced-round AES-128 by 200 to 800 times. Crucially, neither result impacts real-world production systems—full AES-128 and NIST's primary HAWK parameters remain secure. However, the achievement highlights a profound shift: AI is moving beyond coding assistance into high-level mathematical discovery. The research required significant investment—costing roughly $100,000 in API usage per breakthrough and hundreds of hours of human verification, proving that verification remains the ultimate bottleneck as AI takes on complex research. As AI models accelerate scientific and cryptographic exploration, the cybersecurity community must adapt to how we evaluate and reinforce future post-quantum standards. Read more about the findings, technical papers, and reproducibility artifacts on Anthropic’s research blog. #Cybersecurity #ArtificialIntelligence #PostQuantum #Cryptography #AIResearch #Anthropic

An unreleased AI model just did in 60 hours what human cryptographers couldn't crack in years. Anthropic researchers ran a test using Claude Mythos Preview — a model not yet publicly available — against two cryptographic systems. The results: it improved the best-known attack on HAWK, a post-quantum digital-signature candidate under NIST consideration, cutting its effective key strength in half. It also invented a new technique called the "Möbius Bridge" that speeds up attacks on 7-round AES by 200 to 800 times. Total cost to produce both findings: roughly $100,000 in API usage. To be clear — neither result breaks anything in production. HAWK isn't deployed yet, and full-round AES (the version actually protecting your data) remains secure. But both are genuine research-level advances that had eluded human experts for years, and Anthropic has released a new benchmark, CryptanalysisBench, specifically to track how fast this capability keeps growing. Here's why this matters beyond the headline: cryptographic research has always moved at human speed — years of incremental papers before a real breakthrough. If frontier models can now compress that timeline into days, security teams and standards bodies (like NIST) will need to iterate on post-quantum standards much faster than planned, before those weaknesses get found by someone with worse intentions. Worth remembering next time someone tells you AI is "just autocomplete." #AI #Cybersecurity #Cryptography #Anthropic #PostQuantum #TechNews — Sandeep Raiza

An unreleased AI model just did in 60 hours what human cryptographers couldn't crack in years. Anthropic researchers ran a test using Claude Mythos Preview — a model not yet publicly available — against two cryptographic systems. The results: it improved the best-known attack on HAWK, a post-quantum digital-signature candidate under NIST consideration, cutting its effective key strength in half. It also invented a new technique called the "Möbius Bridge" that speeds up attacks on 7-round AES by 200 to 800 times. Total cost to produce both findings: roughly $100,000 in API usage. To be clear — neither result breaks anything in production. HAWK isn't deployed yet, and full-round AES (the version actually protecting your data) remains secure. But both are genuine research-level advances that had eluded human experts for years, and Anthropic has released a new benchmark, CryptanalysisBench, specifically to track how fast this capability keeps growing. Here's why this matters beyond the headline: cryptographic research has always moved at human speed — years of incremental papers before a real breakthrough. If frontier models can now compress that timeline into days, security teams and standards bodies (like NIST) will need to iterate on post-quantum standards much faster than planned, before those weaknesses get found by someone with worse intentions. Worth remembering next time someone tells you AI is "just autocomplete." #AI #Cybersecurity #Cryptography #Anthropic #PostQuantum #TechNews — Sandeep Raiza

🚨 AI is starting to reshape cryptography research faster than many expected. A recent report from The Hacker News highlights how Anthropic's Claude identified new weaknesses in the experimental post-quantum signature scheme HAWK-256 and improved an attack against a reduced version of AES-128. Before jumping to conclusions: ✅ This does not mean that the AES protecting today's Internet has been broken. ✅ It does not mean post-quantum cryptography is suddenly insecure. ✅ It does show that frontier AI models are becoming powerful assistants for cryptanalysis, helping researchers discover vulnerabilities and evaluate cryptographic designs at unprecedented speed. As AI capabilities continue to advance, they are accelerating both defensive research and offensive security analysis. This reinforces an important reality: preparing for the post-quantum era isn't just about quantum computers anymore—it's also about AI. The question is no longer if AI will transform cryptography. It's how quickly it will redefine the way we design, test, and secure cryptographic systems. https://lnkd.in/e7bgjF5u #CyberSecurity #AI #Cryptography #PostQuantum #PQC #Anthropic #Claude #InfoSec #CyberDefens

This one PQC- algorithm, and that should cause speculation. We have NIST approved algorithms that are going to be the standard that tells us what threat exists when Quantum arrives.

‼️ Claude Mythos Preview Autonomously Breaks Post-Quantum Encryption | Source: https://lnkd.in/g8Jza-V5 Anthropic researchers using Claude Mythos Preview have uncovered mathematical flaws in major cryptographic algorithms that human experts failed to spot for years. The AI found improved attacks against HAWK, a post-quantum digital signature scheme, and a reduced-round version of AES, the world’s most widely used symmetric cipher. Neither result threatens production systems today, but both mark a significant shift in how artificial intelligence can stress-test the foundations of digital security. #cybersecuritynews #ClaudeMythos

Claude Mythos Preview Discovers Cryptographic Weaknesses That Human Experts Missed for Years July 28, 2026 Anthropic researchers using Claude Mythos Preview have uncovered mathematical flaws in major cryptographic algorithms that human experts failed to spot for years. The AI found improved attacks against HAWK, a post-quantum digital signature scheme, and a reduced-round version of AES, the world’s most widely used symmetric cipher. Neither result threatens production systems today, but both mark a significant shift in how artificial intelligence can stress-test the foundations of digital security....