Two days before South Korean cybersecurity firm ITCEN PNS announced that its cryptographic library EdgeCrypto v4.2 had cleared the country’s mandatory government certification program, an AI model built by Anthropic discovered a fatal flaw in one of the algorithms being considered as a future standard for exactly the same purpose. The algorithm was called HAWK. By July 29, the day ITCEN PNS publicized its certification, HAWK’s developers had withdrawn it from the National Institute of Standards and Technology’s post-quantum standardization process. EdgeCrypto had not been built with HAWK inside it. It had been built precisely so that it would not need to be.

That timing is not coincidence. It is, rather, a vivid demonstration of why the industry has converged on hybrid post-quantum cryptography — running a classical encryption algorithm and a quantum-resistant one simultaneously — rather than migrating to a single new algorithm and hoping that algorithm holds.

An AI Model Found What Two Years of Human Review Missed

HAWK is a digital signature scheme designed to remain secure against quantum computers, built on a mathematical structure called the Lattice Isomorphism Problem. It had passed two full rounds of expert evaluation in NIST’s competition for additional digital signature standards before Anthropic’s Claude Mythos Preview model sat down with it — semi-autonomously, using a scaffold built on Claude Code — and in approximately 60 hours of work found a mathematical symmetry that every prior human reviewer had missed.

The vulnerability is a previously unexploited property called a nontrivial automorphism in HAWK’s lattice structure. By exploiting this symmetry, Mythos developed an attack that doesn’t break HAWK outright — the attack still takes exponential time — but effectively halves the algorithm’s security level, meaning HAWK would need to double its key sizes to maintain the security margins it was designed to provide. Doubling HAWK’s key sizes would eliminate most of the engineering advantages that made it a competitive candidate.

Google post-quantum cryptography expert Sophie Schmieg said the result effectively ended HAWK’s chances of becoming a standard. Ellen Boehm, senior vice president at digital identity firm Keyfactor, framed the result as the process working correctly: the research, she said, underscores the need for organizations to understand exactly where cryptography lives in their infrastructure — which algorithms, which systems, which business processes depend on them — and to have a PQC readiness plan in place before the next algorithmic collapse arrives.

NIST has emphasized that the HAWK attack does not affect the three finalized post-quantum cryptographic standards — ML-KEM (FIPS 203), ML-DSA (FIPS 204), and SLH-DSA (FIPS 205), finalized in August 2024 — because those standards rest on different mathematical hardness assumptions. ML-KEM and ML-DSA are built on the Module Learning With Errors (MLWE) problem, which involves recovering a secret from a noisy system of linear equations over a polynomial ring. Mythos’s nontrivial automorphism attack is specific to the Lattice Isomorphism Problem underpinning HAWK; there is no known analogue that transfers to MLWE.

But the larger lesson of HAWK is not about a single algorithm. It is about the assumption that the algorithms chosen today will remain secure indefinitely. That assumption has now been publicly invalidated twice in four years — first with SIKE, which was completely broken on a laptop in 2022 after it had advanced to the fourth round of NIST’s main standardization competition; and now with HAWK, broken by an AI model working with a non-expert researcher in fewer than three days of compute. As Anthropic noted in its research disclosure, such critical findings late in the process are not unprecedented — during NIST’s standardization of ML-KEM and ML-DSA, several competing proposals were found to be insecure.

What KCMVP Is and Why It Matters for Anyone Procuring Encryption

The certification that EdgeCrypto v4.2 received is called KCMVP — the Korea Cryptographic Module Validation Program — and it operates as the functional equivalent of the United States’ FIPS 140-3 validation regime, administered by the National Security Research Institute under South Korea’s National Intelligence Service. Under South Korea’s Electronic Government Act, any cryptographic module operating on national administrative networks must hold KCMVP validation — the certification is not a competitive differentiator; it is a condition of entry into the country’s public-sector market.

KCMVP validates that modules comply with the Software Cryptographic Module Validation Standard or its hardware equivalent (KS X ISO/IEC 19790) and that they implement the suite of Korean national algorithms — including ARIA, SEED, LEA, and HIGHT — alongside any internationally recognized standards the module also supports. The evaluation process typically takes approximately 18 months, involving multiple rounds of technical review, supplement requests, and retesting cycles.

The guidelines governing what KCMVP would accept were revised in December 2025, opening the program to post-quantum cryptography hybrid modules for the first time. Exgate, a Korean network security specialist, was the first company through the gate in April 2026, receiving what it described as the country’s first KCMVP certification for a PQC-based software hybrid module. ITCEN PNS’s EdgeCrypto certification, announced July 29, represents the second company to clear the bar since the guidelines changed.

The practical consequence of KCMVP certification is procurement access. Any vendor without it is disqualified from supplying cryptographic products to South Korean government ministries, state enterprises, defense institutions, and other public-sector bodies — a procurement gating function that is structurally identical to the FIPS 140-3 requirement for US federal contracts. South Korea’s Ministry of Science and ICT has simultaneously been expanding its domestic PQC pilot programs in 2026 to cover telecommunications, financial networks, intelligent transportation, defense smart military platforms, and satellite communications, having completed earlier pilots in healthcare, energy, and public administration.

How Hybrid PQC Works and Why the Math of Its Safety Case Changed This Week

EdgeCrypto v4.2 is a software cryptographic library that runs two algorithm tracks simultaneously. One implements the classical algorithms — the RSA and elliptic-curve-based cryptography that enterprise and government systems have relied on for decades. The other implements NIST-standardized post-quantum algorithms: specifically ML-KEM (FIPS 203) for key encapsulation and ML-DSA (FIPS 204) for digital signatures.

The security principle of hybrid deployment is straightforward: a session key established using both a classical algorithm and a post-quantum one is broken only if both families are broken simultaneously. If a quantum computer powerful enough to break RSA or elliptic-curve Diffie-Hellman eventually arrives, the ML-KEM layer still holds. If a novel mathematical attack — of precisely the kind Anthropic’s AI just performed against HAWK — is discovered in the PQC algorithm, the classical layer still holds.

HAWK’s withdrawal makes this safety case concrete in a way that abstract arguments about algorithm risk did not. Anthropic noted in its research disclosure that finding critical flaws late in the standardization process is not unprecedented — during the evaluation of ML-KEM and ML-DSA, several competing proposals were found to be insecure. The 2022 SIKE break — where a laptop running a classical attack demolished a fourth-round NIST candidate — is the most recent prior instance. HAWK is the second in four years, and it was found not by a human team working over months but by an AI model in 60 hours.

The specific difference in mathematical foundations explains why ML-KEM and ML-DSA survive while HAWK did not. ML-KEM and ML-DSA rest on the Module Learning With Errors problem — recovering a secret from a noisy system of linear equations corrupted by small random errors. This problem has no known efficient solution for quantum or classical computers; it lacks the periodic mathematical structure that Shor’s algorithm exploits in RSA and elliptic-curve systems, and it lacks the automorphism structure that Mythos exploited in HAWK. Running ML-KEM alongside classical ECDH in a hybrid implementation means an organization is protected by two independent mathematical hardness assumptions rather than one.

The engineering tradeoffs of hybrid deployment are real: ML-KEM public keys (1,184 bytes for ML-KEM-768) are roughly 37 times larger than an X25519 elliptic-curve key (32 bytes), and ML-DSA signatures (2,420 bytes for ML-DSA-44) dwarf an ECDSA signature’s footprint. In TLS handshakes, this translates to measurable latency increases — studies of post-quantum TLS performance in financial infrastructure have found roughly a 23% increase in connection setup time for high-frequency applications. For most enterprise applications, this is a manageable engineering cost. For high-frequency trading systems, it requires more careful planning.

From Module to Migration Platform: What EdgeCS and CBOM Add

ITCEN PNS has positioned the KCMVP certification not as the end of a product development cycle but as the cryptographic foundation of what it calls a quantum-safe transition platform, pairing EdgeCrypto with a second product called EdgeCS — Edge Cryptography Scanner.

EdgeCS addresses a problem that precedes deployment: most organizations do not know with any precision what encryption they are currently running, where it lives in their infrastructure, or how urgently each implementation needs to be replaced. The tool analyzes cryptographic libraries within running processes, automatically inventories what it finds, and outputs what ITCEN PNS describes as a Cryptographic Bill of Materials, or CBOM. The CBOM concept extends the Software Bill of Materials (SBOM) framework — which tracks software components and their vulnerabilities — into the cryptographic domain specifically, mapping algorithm types, their system locations, and their replacement priority.

Together, the two products form the first two stages of a three-step migration methodology ITCEN PNS describes as Detection, Analysis, and Application. Detection means generating the CBOM via EdgeCS. Analysis means assessing which implementations face the most urgent migration need and in what order to address them. Application means migrating those systems to PQC-compliant encryption using EdgeCrypto as the certified cryptographic engine. CEO Park Won-kyu framed the value proposition directly: “Many companies and institutions recognize the necessity of PQC but are often at a loss as to where to actually begin.”

The broader context for this framing is the gap that enterprise surveys have consistently documented. A DigiCert survey of 1,001 IT and cybersecurity decision-makers across the US, UK, and Australia, published July 23, found that 87% of organizations are planning, testing, or implementing PQC initiatives, but only 7% have deployed quantum-safe or hybrid cryptography broadly — and deployment has grown by just two percentage points in a year. Legacy system complexity has replaced standards uncertainty as the primary blocker; the NIST standards are now settled, but finding and replacing hardcoded cryptographic components across production infrastructure is the engineering challenge that remains.

Korea’s Own PQC Algorithm Enters the Standardization Pipeline

Alongside its commercial migration products, ITCEN PNS has also invested in the upstream standards layer. In June 2026, the company announced its involvement in standardizing NTRU+, a quantum-resistant digital signature algorithm developed in South Korea.

NTRU+ has completed submission of a draft standard for selection as a national standard in South Korea’s broadcasting and telecommunications standards framework and is also progressing through international procedures — an Internet-Draft has been filed with the Crypto Forum Research Group under the Internet Research Task Force. If adopted at the IETF level, NTRU+ would represent a Korean domestic algorithm entering the global pool alongside the NIST-standardized ML-KEM and ML-DSA family. For ITCEN PNS, the goal is straightforward: embed NTRU+ in future EdgeCrypto versions and further expand the diversity of algorithm families the library can hedge across.

This is the same logic applied at the module level — more independently-derived algorithm families, more mathematical independence between failure modes — extended to the standards pipeline itself. HAWK demonstrated that even an algorithm surviving multiple rounds of expert review can carry hidden vulnerabilities. A richer ecosystem of independently-designed candidates provides a more robust set of fallback options if another HAWK-like discovery occurs.

Deployment Horizon and Target Markets

ITCEN PNS has identified four priority verticals for its quantum-safe transition platform: the public sector, finance, defense, and manufacturing — the industries in South Korea where both security requirements and procurement processes are most tightly linked to government certifications like KCMVP. With the EdgeCrypto certification in hand, the library is now eligible for deployment in government networks, financial infrastructure, and industrial control systems across the country.

The international PQC market now operates under hard regulatory deadlines that sit in the near term, not the distant future. In the United States, the NSA’s Commercial National Security Algorithm Suite 2.0 requires all new National Security System acquisitions to support post-quantum algorithms by January 1, 2027. NIST’s own transition guidance (NIST IR 8547) calls for phasing out RSA and elliptic-curve algorithms in federal systems after 2030 and prohibiting them entirely after 2035. September 21, 2026 — less than two months from the date of this article — marks the date when NIST moves all remaining FIPS 140-2 validated certificates to Historical status, after which only FIPS 140-3 validated modules qualify for new US federal procurement. South Korea’s own KCMVP timeline, driven by the government’s five-sector PQC pilot expansion, creates parallel urgency for domestic procurement.

The Anthropic HAWK disclosure adds one more datapoint to that urgency that is not about deadlines at all. HAWK’s collapse is a reminder that the current PQC candidate ecosystem, however carefully designed and however rigorously reviewed, still contains vulnerabilities that human review has not yet found. AI-assisted cryptanalysis will find more of them. Every organization waiting for the standards picture to become fully settled before beginning migration is waiting for a condition that may not arrive in any stable form — because the process of finding and removing algorithmic vulnerabilities, as Anthropic has now demonstrated, is itself being accelerated by the same AI capabilities that make other parts of cybersecurity harder. Hybrid deployment, structured around algorithm-family diversity rather than a single bet on any one standard, is the hedge that survives that process regardless of what comes next.

Frequently Asked QuestionsWhat exactly happened to the HAWK algorithm, and does it affect the PQC standards already in use?

HAWK was a digital signature scheme being evaluated by NIST as a candidate for post-quantum standardization — it had passed two rounds of expert review over approximately two years. On July 28, 2026, Anthropic disclosed that its Claude Mythos Preview AI model had found a previously unknown mathematical symmetry (a nontrivial automorphism) in the lattice structure underpinning HAWK’s security. This symmetry enables an attack that effectively halves HAWK’s security level, meaning its key sizes would need to double to maintain the same protection — a change that would eliminate most of what made HAWK appealing. The HAWK team withdrew the algorithm from the NIST process the following day. Critically, the attack does not affect the three post-quantum standards NIST finalized in August 2024 — ML-KEM (FIPS 203), ML-DSA (FIPS 204), and SLH-DSA (FIPS 205) — because those standards are built on different mathematical hardness assumptions (Module Learning With Errors and hash-based security), which the Mythos attack does not address.

What is KCMVP, and why should organizations outside South Korea care about it?

KCMVP is South Korea’s national cryptographic module validation program, administered by the National Security Research Institute under the National Intelligence Service. It is the functional parallel to FIPS 140-3 in the United States: any cryptographic module deployed in South Korean government, defense, or public-sector institutions must hold KCMVP certification as a legal requirement under the Electronic Government Act. For organizations outside South Korea, KCMVP matters as a signal about how national procurement systems function as forcing mechanisms for PQC adoption — once a government certifies hybrid PQC modules and mandates their use, the market for uncertified classical-only modules in that jurisdiction contracts rapidly. The same dynamic is playing out in the US with FIPS 140-3, whose requirement for new federal procurement takes full effect in late September 2026.

Why does running two algorithms simultaneously protect against a situation like HAWK’s collapse?

Hybrid PQC works by generating a session key that is only broken if both the classical algorithm and the post-quantum algorithm are broken simultaneously. If an attacker with a quantum computer breaks the classical component, the PQC layer still holds. If a cryptanalyst — or an AI — discovers a mathematical weakness in the PQC component, the classical layer still holds. The 2022 collapse of SIKE, a fourth-round NIST candidate, illustrated exactly this principle: Google had been running hybrid TLS tests that paired SIKE with X25519 (an elliptic-curve algorithm). When SIKE was broken, users protected by the hybrid scheme were unaffected, because X25519 still held the session. HAWK is the same scenario, one round later and found by an AI rather than a human researcher. An organization running ML-KEM alongside classical ECDH in a hybrid implementation is not dependent on ML-KEM being unbreakable — it is protected by the mathematical independence of two separate hardness assumptions.

What is a Cryptographic Bill of Materials and how does it differ from a Software Bill of Materials?

A Software Bill of Materials (SBOM) is a structured inventory of the software components used in an application — libraries, dependencies, versions, and their known vulnerabilities. A Cryptographic Bill of Materials (CBOM) applies the same inventory concept specifically to cryptographic assets: what encryption algorithms are running in a given system, what protocols and libraries implement them, where they live in the system architecture, and how urgently each needs to be replaced ahead of quantum threats. A CBOM tells an organization not just that it is running OpenSSL but that a specific version of OpenSSL is performing RSA-2048 key exchange on a particular production server that processes financial records with a 10-year retention requirement — a much more actionable output for migration planning. ITCEN PNS’s EdgeCS generates CBOMs automatically by analyzing cryptographic libraries within running processes.