Post-quantum cryptography (PQC) is a family of algorithms designed to protect communications and data against both classical computers and future large-scale quantum computers. PQC runs on conventional computers; it does not require quantum hardware.
The main concern is that sufficiently capable quantum computers could break some public-key cryptography used today. PQC therefore focuses primarily on quantum-resistant key establishment and digital signatures.
NIST standards
NIST finalized three standards in 2024:
- FIPS 203: ML-KEM for key encapsulation.
- FIPS 204: ML-DSA for digital signatures.
- FIPS 205: SLH-DSA, a stateless hash-based signature scheme.
NIST’s 2026 publications continue work on migration, crypto-agility and additional signature candidates.
Why migrate now?
Migration can take years because organizations must inventory TLS, VPNs, certificates, HSMs, signing services, applications, embedded devices and third-party dependencies.
Long-lived confidential data also creates a “harvest now, decrypt later” concern: encrypted traffic captured today may be targeted later.
Practical example
An Indonesian company storing sensitive customer data for many years can inventory RSA and elliptic-curve dependencies, identify long-lived data and test quantum-resistant options in APIs, VPNs and certificate infrastructure before a production migration.
The goal is not merely replacing one algorithm. It is building crypto-agility: the ability to change cryptographic mechanisms without redesigning the whole system.
Limits
PQC does not fix weak identity management, stolen credentials or insecure software. New algorithms can also change key sizes, signatures, bandwidth and CPU requirements, so interoperability and performance testing remain necessary.
Sources
NIST PQC: https://www.nist.gov/pqc NIST PQC publications: https://csrc.nist.gov/Projects/Post-Quantum-Cryptography/publications NIST FIPS 203/204/205: https://csrc.nist.gov/projects/post-quantum-cryptography
Why PQC matters
Pqc matters when a system has to operate reliably beyond a demonstration or isolated experiment. The important engineering question is not whether a technology sounds advanced, but whether its assumptions, dependencies and failure modes are understood. In this context, the central concepts are cryptographic migration, ML-KEM, ML-DSA, SLH-DSA, crypto-agility. A production implementation should connect those concepts to explicit requirements, measurable outcomes and controls.
For a glossary reader, the most useful distinction is between the concept itself and the surrounding implementation. A protocol does not automatically make an integration secure. A model does not automatically make a decision correct. A cloud region does not automatically establish sovereignty. A security control does not eliminate every threat. Good architecture starts by identifying exactly which problem a mechanism solves and which problems remain outside its scope.
How to evaluate an implementation
A practical evaluation starts with the workload and its constraints. Identify the users, data, interfaces, dependencies, expected scale and consequences of failure. Then define acceptance criteria before selecting a particular vendor, model, algorithm or deployment pattern.
The next step is dependency mapping. Record the components that must remain available for the system to work, including external services, hardware, libraries, identities, data sources and operational personnel. For emerging technology, dependency mapping is particularly important because terminology can hide substantial differences between implementations.
Security should be evaluated as a system property. Consider authentication, authorization, secret handling, isolation, logging, monitoring, update mechanisms and recovery. Where the technology changes how data or commands move, document those flows explicitly.
Performance should also be measured under representative workloads. Useful dimensions can include latency, throughput, memory, compute utilization, bandwidth, error rate and cost. Benchmarking only a best-case example can produce misleading conclusions.
Finally, test failure modes. Ask what happens when a dependency is unavailable, input is malformed, the model is uncertain, a key cannot be released, a tool receives an unsafe request, or a security control is bypassed. A mature implementation has a defined fallback or safe failure mode.
Architecture and lifecycle
The technology should be considered across its lifecycle rather than only at deployment. Requirements change, dependencies are upgraded, data becomes stale, vulnerabilities are disclosed and operational teams change. Versioning and change management are therefore part of the technical design.
A useful lifecycle has five stages: define requirements, build a controlled prototype, evaluate with representative tests, deploy with monitoring, and continuously review the result. Each stage should produce evidence that can be inspected later.
Documentation should record important assumptions. For example, if a system depends on a particular model capability, cryptographic algorithm, cloud jurisdiction, hardware feature or external protocol version, that dependency should be visible to maintainers. Hidden assumptions become technical debt.
Practical decision framework
Use four questions when comparing implementations:
- Capability: Does it solve the intended problem at the required quality?
- Risk: What new failure, security, privacy or compliance risks does it introduce?
- Operational fit: Can the organization monitor, update, troubleshoot and recover it?
- Economics: Is the total cost justified by the value of the successful outcome?
This framework prevents a common mistake: optimizing one technical metric while ignoring the complete system. The cheapest model can produce expensive review work. The fastest architecture can create unacceptable exposure. The strongest isolation can be operationally impractical. The most feature-rich protocol can increase integration complexity.
Example implementation pattern
A production team can start with a narrow workload, establish a baseline using the existing system, and introduce the new technology behind a controlled interface. Instrumentation should capture the metrics that matter before a migration begins. The team can then compare quality, reliability, security and cost against the baseline.
For higher-risk changes, use staged rollout, limited permissions and rollback mechanisms. Keep a record of the version or configuration used in each evaluation. This is especially important for systems whose behavior can change after model, dependency or policy updates.
The practical principle is: inventory algorithms, protocols, certificates, HSMs, applications and third-party dependencies before changing production cryptography. The goal is not to maximize use of the technology, but to apply it where its specific capabilities produce measurable benefit without creating uncontrolled dependencies.
Common mistakes
One common mistake is treating a new technology as a complete solution to a broader problem. Another is copying a reference architecture without checking whether its assumptions match the organization’s workload.
A third mistake is measuring only technical output. Production quality includes operational reliability, security, maintainability, cost and user impact. A fourth is ignoring migration and exit planning. Even when a system works well today, maintainers should know how it will be upgraded, replaced or shut down.
For emerging technologies, terminology also changes quickly. Definitions, specifications and product capabilities should therefore be tied to dated primary sources where possible. A glossary entry should explain the stable concept while clearly identifying time-sensitive implementation details.
Security and governance
Security controls should be proportionate to the impact of failure. Sensitive data, privileged operations and irreversible actions deserve stronger controls than low-risk informational workflows.
Governance should identify owners for the technology, its data, its configuration and its security decisions. Logs should be retained according to applicable requirements, and access should follow least privilege. Where external providers are involved, contracts and service documentation should be part of the dependency assessment.
For regulated or sensitive workloads, technical controls should be mapped to the actual legal and organizational requirements rather than assuming that a technology label constitutes compliance.
What changes as the technology matures?
Early implementations often optimize for capability and experimentation. Mature implementations increasingly emphasize interoperability, lifecycle management, security, observability and predictable economics. This transition is important for readers because a technology can move from experimental terminology into production practice without every implementation becoming equivalent.
The most durable knowledge is therefore the architecture principle: identify the problem, define the trust boundary, measure the outcome, control the dependencies and validate the failure modes. Product names and implementation details can change while those principles remain useful.
FAQ
Is this technology suitable for every organization? No. Suitability depends on workload, risk, maturity, skills, regulatory requirements and economics.
Should an organization adopt it immediately? Not necessarily. A controlled evaluation against a real workload is usually more informative than adoption based on market enthusiasm.
What should be measured first? Establish a baseline for quality, reliability, security and cost, then compare the new implementation against it.
Does using the technology guarantee security or compliance? No. Security and compliance depend on the complete system, configuration, processes and applicable requirements.
How should emerging implementations be maintained? Pin versions where appropriate, monitor upstream changes, document dependencies, retest after upgrades and keep a practical rollback or replacement path.
What is the main lesson? Treat the technology as one component of a larger system. Its value comes from solving a defined problem with measurable benefit and controlled risk.
Further reading
The primary sources listed in this entry should be treated as the authoritative starting point for standards, specifications and regulatory interpretation. Product documentation and implementation guidance should be checked separately because vendor capabilities can change faster than the underlying concept.
Build an inventory before migrating algorithms
Post-quantum cryptography (PQC) migration begins with cryptographic discovery. Identify where public-key encryption, key agreement, digital signatures, certificates, secure boot, VPNs, software signing, device provisioning, and long-lived archives rely on algorithms that may need replacement. Include embedded products and third-party dependencies because a system may contain cryptography that is not visible in the main application repository. Record algorithm, protocol, library, key size, certificate lifetime, data sensitivity, and the person responsible for the dependency.
Prioritize systems using the expected confidentiality lifetime, exposure to captured traffic, migration lead time, and difficulty of updating deployed devices. Data that must remain secret for many years deserves early attention even if the system is not the most visible. A cryptographic inventory should be maintained as software changes rather than treated as a one-time spreadsheet.
Hybrid deployment and interoperability
A transition may combine conventional and post-quantum mechanisms so that compatible participants can establish protection while implementations are evaluated. Hybrid designs can increase message sizes, computational requirements, and protocol complexity, so test certificate chains, handshake limits, network devices, storage formats, and constrained clients. Do not invent an ad hoc combination of algorithms; follow standardized protocol guidance and the security properties it claims.
Testing and governance
The migration plan should track standards status, supported library versions, vendor roadmaps, and the ability to revoke or replace cryptographic components. Test performance on actual target hardware and measure handshake latency, memory, signature size, bandwidth, and compatibility. Software signing and certificate management deserve particular attention because a failed update can leave devices stranded.
PQC protects against a class of future cryptanalytic capability; it does not fix weak authentication, stolen private keys, poor random-number generation, vulnerable endpoints, or implementation flaws. Keep the broader cryptographic hygiene program active, document exceptions, and define dates or events that trigger reassessment. The objective is a controlled transition with evidence of interoperability and security, not simply changing an algorithm name in configuration.