For years, the story around quantum computing and cyber security has been relatively simple: one day, sufficiently capable quantum computers may be able to break much of the public-key cryptography that protects the internet, financial systems, software updates and sensitive communications.
That threat is real. But it is no longer the whole story.
A newer development is beginning to matter:
AI is helping researchers test cryptographic systems faster and more aggressively than before. It is not replacing cryptographers, and it has not “broken post-quantum cryptography.” But it is beginning to change the speed at which weaknesses can be discovered, checked and acted upon.
That should change how we think about the transition to quantum-safe security.
The headline is not that PQC failed
The immediate trigger for this discussion was HAWK, a digital-signature scheme being considered in NIST’s additional post-quantum cryptography process.
Last month, Anthropic reported that its Claude Mythos model helped researchers find a significant structural weakness in HAWK. However, Shortly afterwards, the HAWK team withdrew the scheme from NIST consideration. It will not become a standard or be deployed as one.This story may sound a bit dramatic, but the right interpretation is not “post-quantum cryptography has failed.”
It is almost the opposite.
HAWK was a candidate, not an approved standard already protecting the world’s critical systems. Candidates exist to be scrutinized. They are put in front of experts, competitors, academics and attackers precisely so that weaknesses can be exposed before governments, companies and infrastructure providers build them into systems that may last for decades.
In other words, a weakness being found before deployment is what a healthy standards process is always meant to achieve.
NIST’s established post-quantum standards are not HAWK.
In 2024, it published ML-KEM as its principal key-establishment standard, alongside ML-DSA and SLH-DSA for digital signatures. NIST says the HAWK result does not affect those final standards.
The bigger story is the speed of review
The more important development though, is not simply that HAWK had a weakness. Cryptographic candidates have always been challenged, and some have always failed.
The difference is that an AI system may now help researchers generate ideas, search mathematical structures, test attack paths and connect possibilities at a pace that a small team of specialists could not easily match on its own.
Think of cryptography as designing a lock for an extremely valuable vault. Before installing that lock everywhere, you want the most skilled lock-pickers in the world to examine it from every angle. You want them to look for design shortcuts, hidden weaknesses and unexpected combinations of parts that make the lock easier to open.
Traditionally, that testing has depended on highly specialised human expertise. It still does. But AI can potentially act as a force multiplier:
proposing hypotheses,
conducting repetitive exploration,
searching across a larger space of possible attacks and
helping researchers identify where their attention may be most valuable.
That may be good news for defenders. It also means that flaws can be detected earlier.
But it also means that the old model of cryptographic assurance - where a design survives a few years of expert review and therefore feels settled - may become less reassuring. AI is making it plausible that designs will face sharper scrutiny sooner.
What HAWK actually tells us?
HAWK was designed as a post-quantum digital-signature scheme.
Digital signatures are not mainly about hiding information. They are about proving that a piece of software, message or transaction genuinely came from the claimed sender and has not been altered.
For example, signatures help verify that a software update really came from its developer. They help establish trust in firmware running inside devices. They can be used for identity, secure transactions and long-lived records.
The reported HAWK attack did not produce an instant practical attack against all post-quantum cryptography. It found a mathematical property that substantially reduced the estimated effort required to recover an equivalent secret key for relevant parameter sets. The practical implication was that restoring the intended security strength would require more conservative parameters, which could undermine HAWK’s performance and efficiency advantages.
This is an important lesson in how cryptography works in the real world.
A scheme can look secure enough at one size, fast enough at another and easy enough to implement at a third. Once a new attack is found, the balance may change. Increasing key sizes or other parameters can restore security, but it may also make a system slower, heavier or more difficult to deploy.
Cryptography is not only maths. It is engineering, economics, software design and long-term risk management.
AI has not broken today’s encryption
It is important not to overstate the story.
The AI-assisted work also produced improved attacks on a reduced-round version of AES, the widely used symmetric encryption standard. But that research did not break full AES, which remains the version used in real systems.
Similarly, the HAWK result does not mean ML-KEM, ML-DSA or all lattice-based cryptography is suddenly unsafe. HAWK was a separate candidate with its own design choices, and it has been removed from consideration before standardisation.
That distinction matters because irresponsible headlines can create the wrong kind of urgency. Organisations do need to prepare for post-quantum migration, but they should not rush into poorly understood changes because an AI-related claim appears on social media.
The better takeaway is this:
AI is not breaking deployed post-quantum standards today. It is making the process of finding and validating weaknesses faster.
That is a serious development, but it is not a reason for panic.
There are now two clocks
Most organisations think about the quantum threat using one clock: when will a quantum computer be capable enough to break RSA or elliptic-curve cryptography?
That is the quantum clock. It matters because much of the internet and enterprise technology still relies on public-key cryptography that was not designed to withstand a large fault-tolerant quantum computer.
But there is now a second clock: the AI clock.
The AI clock is not about a future machine that suddenly breaks encryption overnight. It is about the growing ability to test cryptographic assumptions more quickly, identify hidden weaknesses earlier and reduce the time between a new design being proposed and its security claims being challenged.
The quantum clock explains why organisations will eventually have to migrate.
The AI clock explains why they should not assume that the security assumptions made today will remain untested for years.
Together, those clocks point toward the same conclusion: build systems that can adapt.
The hidden problem is not just TLS
When executives hear “quantum-safe encryption,” they often think about a website using HTTPS. That is only a small part of the challenge.
A company may be able to upgrade the cryptography at its public web edge relatively quickly, particularly when major cloud providers and content-delivery networks support hybrid post-quantum connections. But the harder problem is often inside the organisation.
Where are the signing keys used to approve software releases?
How are internal services authenticating one another?
What secures firmware updates to devices in the field?
What protects encrypted backups that may need to remain confidential for years?
Which identity systems, industrial gateways, VPNs, APIs, containers and machine-to-machine connections rely on older public-key algorithms?
Many organisations cannot answer those questions with confidence.
That is the real operational risk.
You cannot migrate cryptography you have not located. You cannot prioritize critical systems if you do not understand which secrets must remain safe for ten years, twenty years or longer.
Start with a cryptographic inventory
The most useful response to this moment is not to immediately replace every encryption algorithm. It is to establish visibility.
A practical cryptographic inventory should identify:
Where public-key cryptography is used across products, IT systems and operational technology.
Which algorithms, key sizes, libraries and certificates are in use.
Which systems rely on digital signatures, including code-signing, device identity and firmware updates.
Which data must remain confidential for a long time.
Which systems are difficult or expensive to update.
Which suppliers and platforms already offer support for hybrid or post-quantum approaches.
This does not have to begin as a massive enterprise-wide project. A company can start with the systems where the consequences of failure would be greatest: customer data, software release pipelines, identity infrastructure, high-value intellectual property, industrial control systems and long-life connected devices.
For a startup, this can be even more important.
Early architecture decisions become embedded surprisingly quickly. A product designed today may still be in service a decade from now. If its security model cannot evolve, the eventual migration may become a costly retrofit.
Crypto-agility is the real objective
The phrase “cryptographic agility” sounds technical, but the basic idea is straightforward.
A cryptographically agile system can change its security methods without requiring a complete redesign of the product, network or service.
That does not mean swapping algorithms casually. Cryptographic changes should be carefully tested and independently reviewed. But it does mean avoiding systems where one old algorithm is hard-coded into every component and cannot be replaced without disrupting the whole business.
For example, an organisation should be able to identify where certificates are issued, which libraries handle signatures, how keys are stored, and what must change when an algorithm needs to be retired.
This is where the HAWK story becomes useful beyond cryptography research.
A candidate scheme was challenged. A weakness was found. The candidate was withdrawn. That is what adaptation looks like at the standards level.
Organisations need the same ability at the deployment level: to change course before an avoidable technical dependency becomes a security crisis.
Quantum computing sets the long-term deadline for migration. AI-assisted cryptanalysis may be bringing forward the moment when weak assumptions are exposed.
The encouraging part of HAWK story is that the system worked.
Researchers investigated a candidate. A weakness was identified. The designers and standards community acted before the scheme was standardized or broadly deployed. NIST’s selected standards remain in place, while the broader PQC process continues to examine additional options.
That is not a sign of chaos. It is the value of open scrutiny.
The uncomfortable part however, is that AI may make that scrutiny more intense. It may force cryptographers, standards bodies, vendors and security teams to operate at a faster pace. The bottleneck may increasingly be not just finding a weakness, but validating it, responsibly disclosing it, updating software and communicating clearly about real-world impact.
That will require more expertise, not less and a sensible response is neither complacency nor alarm.
Do not conclude that AI has broken post-quantum cryptography. It has not.
Do not assume that quantum risk can be ignored until a future breakthrough in quantum hardware. It cannot.
Instead, treat this as a prompt to do the essential work:
Map where your organisation depends on cryptography.
Identify long-lived data, software signatures and trust anchors.
Ask vendors about post-quantum roadmaps and hybrid deployment options.
Build systems that can support future algorithm changes.
Make cryptographic agility a normal engineering requirement, rather than an emergency project.