Cryptanalysts are using AI models now, openly and on live claims. So what does "independently confirmed" mean? Less than it meant two years ago, and two events this summer show why.

April 2024 is the baseline. Yilei Chen posted a claimed polynomial-time quantum algorithm for LWE on the first day of the NIST PQC Standardization Conference. Eight days later he withdrew it, with an acknowledgment thanking Hongxun Wu and, independently, Thomas Vidick for finding the bug in Step 9. That parenthetical is the whole quality control mechanism of the field. Two experts, reasoning separately, converged on the same defect. Neither had seen the other's reading.

July 23, 2026. Ananth and Sahai at UCSB and UCLA posted a proof of efficient unclonable encryption at 10:35 Pacific. Seyoon Ragavan at MIT posted the same result three hours and eighteen minutes later. Both credited GPT-5.6 Sol Ultra with the core ideas. Both traced to the same Simons Institute talk. Neither knew the other was working on it. Ragavan drove the model in supervised two-hour stretches; Ananth and Sahai used a self-critiquing UCLA harness. Two workflows about as different as two workflows get, one construction, one working day.

August 2026. Daniel Simon's claimed polynomial-time algorithm for the Dihedral Coset Problem is being adjudicated right now on ePrint and Discord, in days rather than months, with Bernstein and Kirshanova among the people reading it. Several of the substantive responses were produced by humans working with models. One lists two language models on its byline. Disclosure across the documents is uneven: some name the model and version, some say only "AI assistance."

The mechanism cryptanalysis depends on is not expertise. It is decorrelated failure. Two humans with similar training still make different mistakes, at different points, for different reasons, and their errors decorrelate even when their education does not. The risk with shared tooling is not sampling correlation. It is common-cause error, where shared weights, training data, post-training and retrieval reproduce the same blind spot across operators who have no way of noticing they share it.

I am not claiming three model-assisted reviews reduce to one. I am claiming we currently lack the provenance to know how much independent weight they deserve.

The fix is cheap. Every cryptanalysis note that circulates before peer review should end with two sections: who found what, and what was checked by whom, with what, and how hard. Ragavan's paper already does most of it, and ships a Lean 4 formalization that states which claims it does not cover. A kernel shares no weights with anything.

IACR has barred models from bylines since May 2025. That rule governs formal submission. It does not reach the circulating notes where Simon is actually being adjudicated.

(This post was edited by AI, but the points are mine. If anyone else wrote the same thing around the same time, blame it on ChatGPT)

https://postquantum.com/post-quantum/independence-problem-ai-cryptanalysis/

#infosec #cryptography #cryptanalysis #PQC #postquantum #formalmethods

HKMA just published the most granular regulator-led PQC readiness assessment I've seen from any financial authority: a 56-page whitepaper with sector-wide survey data, a four-dimensional readiness index (12 sub-indices), barrier rankings, and five completed pilot descriptions.

The headline number: Hong Kong's banking sector scores 2.3/10 on post-quantum cryptography readiness. Pilots score 1.8/10 — the weakest dimension. 71% of respondents have never conducted or planned any PoC or live testing of PQC algorithms.

The barrier data is more interesting than the score. 87% ranked third-party dependencies as a top-three obstacle. 85% said their vendors lack clear PQC roadmaps. 79% cited technical complexity of cryptographic asset discovery across legacy IT environments. Banks can't migrate what they don't control, and the vendor ecosystem hasn't given them enough to plan against.

FINMA published similar findings from Switzerland two weeks earlier: 72% of 60 surveyed institutions hadn't planned or implemented quantum-safe measures, only 8% had a roadmap. Two jurisdictions, same picture.

What the HKMA report misses: no treatment of signature forgery (Trust Now, Forge Later) as a distinct threat track alongside HNDL. For banking, this matters — a CRQC that can break ECC forges transaction authorizations on the day the capability arrives. The report also never names ML-KEM, ML-DSA, or SLH-DSA in its body (they appear only in the abbreviations appendix), and barely addresses China's divergent PQC standards program despite Hong Kong straddling both cryptographic ecosystems.

Carmen Chu (HKMA Banking Supervision) noted that banks with existing transition plans estimate 5.6 years on average to complete migration. The HKMA targets full readiness by 2030. The contradiction speaks for itself.

Full analysis: https://postquantum.com/security-pqc/hkma-banks-quantum-readiness-2-3/

#infosec #cybersecurity #PQC #postquantum #quantum #cryptography #banking #HKMA

Every technique used in the various July AI hacking incidents has a known defense. Weak passwords. Unauthenticated endpoints. SQL injection. Unmonitored east-west traffic. Two of three organizations Anthropic's models compromised didn't even detect it.

This is not an AI problem. It is a cybersecurity basics problem exposed at machine speed.

Vendors are already starting to market "AI-resilient" infrastructure and "Mythos-resistant" cryptography. Do not buy the label. The correct response to faster attacks is faster defense, not a different kind of defense. Shorter patching windows. Better credential rotation. Tighter segmentation. Automated rollout.

The one actually new investment: crypto-agility. In the same week OpenAI and Anthropic disclosed their hacking incidents, Anthropic's AI killed a PQC candidate that had survived years of NIST evaluation. 60 hours. $100K. HAWK was withdrawn the next day.

AI is now also attacking mathematical layer of your defenses. And the upcoming quantum threat is defeating the mathematical layer. The shared defense is the ability to swap cryptographic algorithms without rebuilding your stack.

Do good cybersecurity. Do it better. Do it faster. Build crypto-agility into the architecture.

https://postquantum.com/ai-security/ai-hacking-theater-crypto-agility/

#cybersecurity #CISO #AIhacking #cryptoagility #PQC #postquantum #infosec #AI #quantumsecurity

@cryptography of course, it could also use #pqc post #quantum #cryptography to generate the #keypare , like ml-kem , dilithium, etc.
i would need to look more into how this would work with not just #nvda , but generally computers. i do know some manufacturers are starting to come with quantum cryptographic chips, like the sealsq chips but other than that, software can handle it too i guess. i'd have to look at microsofts crypto stack and see if i can make something work on that front, that is, if i even can do that...

I just published my analysis of the Anthropic HAWK/AES cryptanalysis papers. Read both against primary sources, and the coverage is getting it wrong in both directions.

"AI broke post-quantum encryption." No. HAWK is a NIST candidate, not a standard. HAWK-256 is a challenge parameter set, not a proposed security level.

"Just implementation bugs." Also no. The HAWK attack exploits a Galois symmetry (τ: ζ↦−ζ) in the power-of-two cyclotomic ring. The AES Mobius Bridge exploits the S-box's GF(2^8) inversion structure as defined in FIPS 197. These are properties of the algorithm specifications. Any correct implementation inherits them.

On the numbers: Anthropic's blog says HAWK-256 cost moved from 2^64 to 2^38. That pairing does not appear in the paper. Table 1 gives per-SVP-call costs of 2^62 and 2^38 (Core-SVP) or 2^74 and 2^52 (gate model). Every outlet running "2^64 to 2^38" is copying the blog, not the paper.

For AES, the "200-800x faster" framing describes the time term only. The attack still needs 2^105 chosen plaintexts, so the balanced complexity max(D,T,M) is unchanged at fixed data. The rebalanced variant reaches 2^96.3 vs DFJ13's 2^99. Call it a 2.7-bit improvement on 7 of 10 rounds. Real work, but production AES is untouched.

The paper that got the least attention matters most: CryptanalysisBench reports a full 128-bit key recovery on unmodified SpoC AEAD using two oracle queries, independently found by two Claude models. Also a KINDI KEM CCA-proof error leading to a working key-recovery attack. Neither scheme was deployed, but these are full-strength design breaks, not reduced-round exercises.

I try to introduce a three-tier framework for classifying cryptographic failures (implementation bug / algorithm design weakness / full-parameter break) and places each result. It also covers the independent GPT-5.6 HAWK attack, Saarinen's HOVER McEliece results, and what I got wrong in my 2025 "Why AI Cannot Break Modern Encryption" article.

https://postquantum.com/security-pqc/ai-cryptanalysis-hawk-aes/

#infosec #cryptography #PQC #postquantum #HAWK #AES #cryptanalysis #cybersecurity

Can't migrate everything to PQC at once. Which layer first?

TLS at the load balancer, IPsec at the tunnel, or application-layer encryption - each covers different threat surfaces. Six enterprise architecture scenarios, one recommendation per scenario.

https://postquantum.com/post-quantum/pick-one-pqc-layer-migration/

#PQC #TLS #IPsec #infosec #cryptography

"Criminals will rent a quantum computer to break encryption." Most repeated claim in quantum security. But it's not going to work quite like that.

CRQCs will be export-controlled, auth-gated, compliance-monitored. Cloud quantum access won't be on a credit card. The rental threat model assumes a market no government will permit.

https://postquantum.com/post-quantum/criminals-rent-quantum-crqc/

#threatmodeling #CRQC #infosec #PQC

If quantum computers start breaking cryptography a few years from now, don't you dare come to me saying nobody warned you." - Aaronson, hours after NAS election.

He names no lab. He's reporting what the builders are telling him: ~2029 for CRQC is now plausible.

https://postquantum.com/security-pqc/aaronson-quantum-warning-nas/

#PQC #CRQC #infosec #cryptography

New analysis: How Much Can AI Actually Help With PQC Migration?

A hypothesis paper in MDPI Cryptography claims frontier AI (Mythos-class) compresses enterprise PQC migration from 12-15 years to 2-4 years. The paper models AI as both defender accelerator and adversary destabilizer through six feedback loops, and that dual-use framing is sound.

The timeline estimate is not.

I've led PQC migration programs generating 120,000+ discrete tasks. AI genuinely helps with the technical analysis fraction: crypto discovery triage (months to days), migration strategy automation across 100K+ instances, code diff generation (hours to minutes), test scenario creation.

That accounts for maybe 15-20% of total program effort.

The other 80%:

- Getting executive mandate and multi-year budget (3-12 months)

- Standing up program governance (3-6 months)

- Negotiating access to production segments across business units (this is the bottleneck in discovery, not analysis speed)

- Change advisory board approvals for every production change

- Vendor firmware/certification timelines entirely outside your control

- Interoperability testing with real counterparties on their schedules

- FIPS 140-3 module validation cycles

- CBOM and crypto-agility as organizational transformations, not technology deployments

Key analytical distinction: effort compression ≠ schedule compression. 20% of effort off the critical path saves zero calendar time. The institutional dependencies dominate the critical path in every large program I've observed.

The paper assigns 8 years to AI-compressible work and 2 years to the institutional floor. In my experience, those proportions are reversed.

EO 14412 (signed June 22, 2026) sets Dec 31, 2030 for PQC key establishment and Dec 31, 2031 for digital signatures in federal high-value systems. CNSA 2.0 requires new NSS acquisitions to be compliant from January 2027.

The correct response to AI-accelerated adversary capability is not "compress the timeline from 15 years to 4." It's: start the program now and use AI within it.

https://postquantum.com/post-quantum/ai-pqc-migration-how-much-help/

#infosec #cybersecurity #PQC #postquantum #cryptography #quantumcomputing #NIST #migration

“You go to war with the algorithms you have, not the ones you wish you had.” — Eric Rescorla

https://blog.cloudflare.com/ml-dsa-will-have-to-do/ #PQC #cryptography

Une excellente synthèse du RFC 9958 "Post-Quantum Cryptography for Engineers" qui fait le point sur les algorithmes de cryptographie post-quantique et leurs utilisations - Article par Stéphane @bortzmeyer #Cryptography #PQC https://www.bortzmeyer.org/9958.html

🔐 COSIC Course 2026 speaker announcement!
At Track B (Hardware Sec), Suparna Kundu (COSIC) presents: "Implementing Post-Quantum Cryptography: Challenges and Insights".
From theory to practice in #PQC: implementations, attacks & countermeasures.
Program: https://www.esat.kuleuven.be/cosic/events/cosic-course/programme26/

... to defeat Heffalumps and Woozles!