Emerging and Future Cryptography

Introduction

Cryptography is entering a period of transition. For decades, public-key systems such as RSA and elliptic-curve cryptography have protected websites, virtual private networks, digital signatures, software updates and financial transactions. Their security depends largely on the difficulty of integer factorisation and the discrete logarithm problem. A sufficiently capable quantum computer could solve these problems efficiently using Shor’s algorithm. At the same time, new computing environments, artificial intelligence, connected devices and privacy regulations are changing what secure information processing must achieve.

Future cryptography is therefore not a single replacement technology. It is an engineering portfolio that includes post-quantum cryptography, quantum-resistant migration, lightweight algorithms, confidential computing, homomorphic encryption, secure multiparty computation and other privacy-enhancing technologies. These methods differ in their assumptions, performance, maturity and appropriate use cases.

This chapter examines these developments at diploma level, with emphasis on practical deployment, risk assessment and professional decision-making.

Learning Objectives

By the end of this chapter, students should be able to:

1. Explain why quantum computing threatens current public-key cryptography.
2. Compare major post-quantum cryptographic families and their trade-offs.
3. Design a practical migration plan for quantum-resistant security.
4. Describe lightweight cryptography and its use in constrained devices.
5. Explain confidential computing, homomorphic encryption and secure multiparty computation.
6. Identify privacy-enhancing technologies and select suitable applications.
7. Assess AI-assisted attacks and controls for protecting AI systems.
8. Evaluate future cryptographic requirements from technical, ethical and governance perspectives.

1. The Quantum Threat and the Need for Transition

1.1 Classical cryptographic assumptions

A cryptographic system is secure only while its underlying assumptions remain valid. RSA relies on the difficulty of factoring a large composite integer. Elliptic-curve systems rely on the difficulty of solving the elliptic-curve discrete logarithm problem. Symmetric encryption, such as AES, does not depend on these public-key problems and is affected differently by quantum computation.

A quantum computer can use Shor’s algorithm to solve integer factorisation and discrete logarithms in polynomial time. This does not mean that an attacker can immediately break existing systems: large, fault-tolerant quantum computers do not currently exist at the scale required for widespread cryptanalysis. However, confidential information may be collected today and decrypted later. This is called a “harvest now, decrypt later” threat. It is particularly serious for government records, health data, trade secrets and long-lived identity credentials.

Grover’s algorithm provides a quadratic speed-up for exhaustive search against a symmetric key. In simplified terms, a key of n bits may offer approximately n/2 bits of quantum search security, assuming the attack conditions of Grover’s algorithm apply. Consequently, moving from AES-128 to AES-256 is often considered a prudent long-term measure, although implementation, key management and operational controls remain essential.

1.2 Security strength and cryptographic agility

Security strength is an estimate of the computational effort required to defeat a scheme. It is not identical to key length. A 256-bit elliptic-curve key does not provide 256 bits of security, and a post-quantum public key may be many times larger than a classical public key while providing a comparable security level.

Cryptographic agility is the ability to replace algorithms, parameters and protocols without redesigning an entire system. It requires:

• algorithm identifiers that are not hard-coded into applications;
• support for versioning and negotiation;
• inventory of certificates, keys and cryptographic libraries;
• separation of cryptographic policy from business logic;
• tested rollback and emergency replacement procedures; and
• monitoring for deprecated or unauthorised algorithms.

Agility is a governance and architecture property, not merely a feature of a cryptographic library.

2. Post-Quantum Cryptography

Post-quantum cryptography (PQC) uses mathematical problems believed to resist both classical and quantum attacks. It normally runs on conventional computers and networks; it does not require a quantum computer.

2.1 Major algorithm families

Standards development has shown that algorithm selection must be based on public analysis rather than novelty. A scheme may be mathematically elegant yet unsuitable because of weak parameter choices, side-channel leakage, poor implementation quality or operational complexity.

2.2 Key encapsulation and digital signatures

A key encapsulation mechanism (KEM) allows two parties to establish a shared secret over an insecure channel. The recipient publishes a public key. The sender encapsulates a randomly generated secret using that public key and sends a ciphertext. The recipient decapsulates it with the private key. The shared secret is then used with symmetric authenticated encryption.

A digital signature provides authenticity and integrity. A signer uses a private key to produce a signature over a message or message digest. A verifier uses the public key to determine whether the message was signed by the expected entity and whether it has been altered.

PQC deployment therefore usually involves replacing both public-key key establishment and public-key signatures. Replacing only one leaves a significant security dependency on the other.

2.3 Hybrid deployment

A hybrid construction combines a classical and a post-quantum mechanism. For example, a session key may be derived from two independently established secrets:

Here, K is the resulting session key, KDF is a secure key-derivation function, ∥ denotes concatenation, and context contains protocol information such as identities and transcript data. The construction assumes that the KDF is secure and that protocol binding prevents an attacker from substituting one component or confusing algorithm identities.

Hybrid operation can reduce transition risk because communication remains protected if one component is later found weak, provided the combination is designed correctly. It also increases message sizes, computation and interoperability complexity. Hybrid security must not be assumed automatically; the protocol must define how failures, downgrade attempts and authentication are handled.

3. Planning a Post-Quantum Migration

A migration programme should begin before a final replacement is needed. The following lifecycle is suitable for many organisations.

Step 1: Discover and classify

Create an inventory of:

• public-key algorithms and key sizes;
• certificates, trust stores and hardware security modules;
• TLS, VPN, email, identity, backup and signing systems;
• cryptographic libraries and embedded firmware;
• data retention periods and confidentiality lifetimes;
• suppliers and external protocols; and
• systems that cannot be upgraded easily.

Classify information according to confidentiality lifetime. A secret that must remain confidential for 20 years deserves earlier action than a short-lived session token.

Step 2: Analyse dependencies and risk

Risk may be prioritised conceptually as:

Likelihood and impact are qualitative or quantitative organisational ratings; they are not probabilities unless explicitly defined as such. A useful prioritisation also considers exposure duration, migration difficulty and the value of the protected information.

Step 3: Test candidates

Test candidate algorithms for CPU use, memory, bandwidth, latency, key and signature sizes, certificate compatibility, hardware support and failure behaviour. Test both normal and adverse conditions, including malformed ciphertexts, repeated negotiations and downgrade attempts.

Step 4: Deploy in controlled stages

Begin with laboratories and non-critical services, then use hybrid protocols for selected production systems. Maintain telemetry for negotiation failures, algorithm use and performance. Avoid disabling legacy algorithms globally until dependent systems have been identified.

Step 5: Govern and review

Define approved algorithms, minimum security levels, rotation periods, ownership and exception processes. Migration is complete only when obsolete algorithms are removed, not merely when new algorithms are available.

4. Lightweight Cryptography

The Internet of Things includes sensors, smart cards, industrial controllers and medical devices with limited memory, energy, processing power and bandwidth. Conventional algorithms may be secure but operationally unsuitable in these environments.

Lightweight cryptography seeks an appropriate balance among security, area, energy, latency, code size and implementation simplicity. It does not mean “weak cryptography.” A lightweight authenticated-encryption scheme should provide confidentiality, integrity and authentication while resisting nonce misuse, forgery and implementation attacks within its intended model.

Important design considerations include:

• Nonce management: A nonce should normally be unique for a given key. Reuse can reveal relationships between plaintexts or completely undermine authenticated encryption, depending on the scheme.
• Device lifecycle: Keys must be provisioned, rotated, revoked and destroyed securely.
• Physical exposure: Devices may be accessible to attackers, making side-channel and fault-injection resistance important.
• Availability: Battery-powered systems may be vulnerable to energy-exhaustion attacks.
• Update capability: A device without secure firmware updates may remain vulnerable for its entire lifetime.

The NIST lightweight cryptography standardisation effort selected Ascon as a standard family for suitable constrained applications. Selection still depends on the device, implementation and threat model; a standard algorithm does not compensate for poor key management.

5. Confidential Computing

Confidential computing protects data while it is being processed, complementing encryption at rest and in transit. It uses a trusted execution environment (TEE), such as an isolated hardware-backed region, to restrict access by the operating system, hypervisor or cloud administrator.

A typical confidential-computing workflow is:

1. A workload is placed inside an isolated execution environment.
2. The environment obtains a cryptographic measurement of its code and configuration.
3. Remote attestation allows a relying party to verify that measurement.
4. Keys or sensitive data are released only if the attestation satisfies policy.
5. The workload processes data inside the protected environment.

The trusted computing base must be kept small. Attestation proves that a particular environment claims to run a particular measurement; it does not prove that the software is free of all vulnerabilities. Side channels, malicious inputs, compromised dependencies, denial of service and flawed policies remain possible.

Confidential computing is useful for cloud analytics, protection of encryption keys, federated learning and processing data shared between organisations that do not fully trust the hosting provider.

6. Homomorphic Encryption

Homomorphic encryption allows computation on ciphertexts such that decrypting the result produces the corresponding computation on plaintexts. In abstract form, an encryption scheme may support:

Here, Enc is encryption, Dec is decryption, f is an allowed function and Eval performs computation over ciphertexts. Correctness assumes valid keys, supported parameters and an implementation that has not exceeded its noise or computational limits.

Types include:

• Partially homomorphic encryption: supports an operation, such as addition or multiplication, repeatedly or in a restricted form.
• Somewhat homomorphic encryption: supports limited combinations of operations.
• Fully homomorphic encryption: supports arbitrary computable functions in principle.

The principal advantage is that a service provider can process encrypted data without seeing the plaintext. The disadvantages are computational cost, ciphertext expansion, parameter selection and programming complexity. Practical uses include privacy-preserving statistics, genomic analysis and outsourced machine learning, but many workloads remain too expensive for real-time general-purpose processing.

7. Secure Multiparty Computation

Secure multiparty computation (MPC) enables several parties to compute a function while limiting disclosure of their private inputs. For parties P₁ through Pₙ with inputs x₁ through xₙ, the goal is that each party learns the permitted output:

without learning the other inputs beyond what y itself reveals. Security depends on the adversary model, including whether parties are semi-honest or malicious, how many may collude and whether communication channels are authenticated.

MPC can support joint fraud detection between banks, collaborative medical research and privacy-preserving salary or credit calculations. It commonly uses secret sharing, oblivious transfer, garbled circuits or arithmetic protocols. Communication rounds and bandwidth can be substantial, especially when malicious behaviour must be detected.

MPC is not automatically private: the output may reveal sensitive information, repeated queries may enable inference, and colluding parties may exceed the assumed threshold. Output controls and statistical privacy methods may therefore be needed.

8. Privacy-Enhancing Technologies

Privacy-enhancing technologies (PETs) reduce unnecessary exposure of personal information while allowing legitimate processing. Relevant techniques include:

• Data minimisation: collect and retain only what is necessary.
• Pseudonymisation: replace direct identifiers with controlled references. It is not anonymisation if re-identification remains possible.
• Differential privacy: add carefully calibrated randomness so that the effect of one individual’s data on an output is limited.
• Federated learning: train models near data sources while sharing updates rather than raw records.
• Private set intersection: determine overlap between datasets without disclosing non-overlapping elements.
• Zero-knowledge proofs: demonstrate that a statement is true without revealing the underlying secret.
• Tokenisation: substitute sensitive values with tokens whose protected mapping is kept separately.

A commonly used differential-privacy parameter is ε (epsilon). Smaller ε generally indicates stronger privacy protection, but privacy depends on the complete mechanism, composition across multiple releases, data distribution and the definition of adjacency. Differential privacy is not a guarantee that every output is harmless.

PET selection should begin with the purpose and threat model. Combining techniques may produce stronger protection, but it may also increase cost, reduce accuracy or create difficult governance questions.

9. AI-Assisted Attacks and Protection of AI Systems

Artificial intelligence can increase the scale and quality of attacks. Examples include automated phishing generation, multilingual social engineering, malware variation, password-guessing prioritisation, deepfake impersonation and rapid analysis of leaked data. AI can also search code for cryptographic misuse, such as hard-coded keys, insecure random-number generation or nonce reuse.

Defensive measures include phishing-resistant authentication, strong email controls, secure software development, threat intelligence, rate limiting, behavioural monitoring and human verification for high-impact transactions. Security teams should assume that convincing language or audio is no longer evidence of identity.

AI systems introduce additional security problems:

• prompt injection and instruction manipulation;
• training-data poisoning;
• model extraction and theft;
• membership inference and sensitive-data memorisation;
• insecure tools and excessive agent permissions;
• adversarial examples and evasion; and
• supply-chain risks in models, datasets and plugins.

Protection requires least privilege for AI agents, isolation of tools, input and output filtering, provenance controls, secrets management, logging, red-team testing and human approval for consequential actions. Cryptography supports these controls through signed model artifacts, authenticated data pipelines, encrypted storage, secure key management and attestation of execution environments. Cryptography cannot by itself determine whether a model’s answer is fair, accurate or safe.

10. Integrated Case Study: A Hospital Preparing for the Quantum Transition

Background

This illustrative case concerns a regional hospital network that operates electronic health records, connected diagnostic devices, cloud backups and an identity federation. Patient records must remain confidential for many years. The network currently uses classical TLS, RSA certificates and elliptic-curve signatures.

Problem

The security team discovers that backup archives are retained for 15 years and that several medical devices cannot be upgraded without vendor involvement. The organisation also plans to use an AI service to analyse medical images. It must prepare for quantum threats without interrupting clinical operations.

Analysis

The highest risks are long-lived records, certificate infrastructure, vendor dependencies and the AI service’s access to sensitive data. The hospital also identifies risks from ransomware, insider misuse, cloud administrators and compromised devices. A single algorithm replacement would not address all of these threats.

Approach

The hospital takes the following actions:

1. It inventories algorithms, certificates, keys, protocols, devices and data-retention periods.
2. It classifies records by confidentiality lifetime and identifies systems requiring early migration.
3. It upgrades symmetric protection for long-lived archives where appropriate and improves key rotation.
4. It tests hybrid key establishment in internal APIs and VPN gateways.
5. It requires vendors to publish PQC transition plans and secure-firmware-update capabilities.
6. It uses confidential computing for selected cloud image-processing workloads, with attestation before releasing keys.
7. It applies data minimisation and pseudonymisation before images leave the clinical environment.
8. It evaluates whether federated learning or MPC can reduce disclosure between hospitals.
9. It signs AI models and datasets, restricts agent permissions and requires human review of clinical decisions.
10. It establishes an algorithm-agility standard and conducts annual migration exercises.

Outcome

The hospital does not claim to be “quantum-proof.” Instead, it reduces exposure, identifies unreplaceable dependencies, tests realistic alternatives and creates a controlled path for future upgrades. Clinical availability is preserved because migration is staged and rollback procedures are tested.

Learning Points

The case demonstrates that future cryptography is an enterprise architecture problem. It combines cryptographic engineering, procurement, privacy, clinical safety, vendor management and governance. The most valuable early action is often discovery and classification rather than immediate algorithm replacement.

11. Challenges, Ethics and Regulation

Future cryptography creates difficult choices. Strong privacy can make lawful investigation more difficult, while exceptional access mechanisms can weaken security for everyone. Organisations must distinguish lawful, targeted procedures from universal backdoors that create systemic vulnerabilities.

Key governance responsibilities include:

• documenting the purpose and legal basis for processing personal data;
• applying data minimisation and retention limits;
• assessing cross-border transfers and supplier access;
• protecting vulnerable populations from discriminatory automated decisions;
• preserving auditability without collecting excessive personal data;
• managing cryptographic keys under separation of duties; and
• reporting vulnerabilities responsibly.

Regulatory compliance is not equivalent to security. A system may satisfy a checklist while using weak architecture, excessive privileges or untested cryptographic dependencies. Conversely, a technically strong design may fail if it does not meet legal requirements for access, retention or accountability.

12. Future Industry Requirements and Best Practices

Cybersecurity professionals will increasingly need the ability to:

1. maintain cryptographic asset inventories;
2. understand both classical and post-quantum security assumptions;
3. test performance and interoperability rather than trusting labels;
4. engineer algorithm agility into applications and protocols;
5. evaluate vendors, standards and software supply chains;
6. protect keys through hardware-backed and policy-controlled systems;
7. combine cryptography with identity, monitoring and secure operations;
8. assess privacy, safety and fairness implications;
9. communicate residual risk to managers and non-specialists; and
10. design for replacement throughout the system lifecycle.

A practical decision framework asks four questions: What must be protected? From whom and for how long? Which computation or disclosure is necessary? What security and privacy assumptions can the organisation realistically maintain? These questions prevent fashionable technologies from being adopted without a clear purpose.

Chapter Summary

Quantum computing threatens RSA and elliptic-curve cryptography through Shor’s algorithm, while Grover’s algorithm reduces the effective security of symmetric exhaustive search. Post-quantum cryptography provides conventional-computer algorithms based on alternative mathematical problems, including lattice-, hash- and code-based constructions. Migration requires discovery, risk classification, testing, hybrid deployment, cryptographic agility and governance.

Lightweight cryptography addresses constrained devices but still requires secure nonce use, lifecycle management and update mechanisms. Confidential computing protects data during processing through hardware-backed isolation and attestation, although it does not eliminate software or side-channel risks. Homomorphic encryption permits selected computation over ciphertexts, while secure multiparty computation allows parties to compute jointly without directly sharing inputs.

Privacy-enhancing technologies include differential privacy, federated learning, private set intersection, zero-knowledge proofs and pseudonymisation. AI expands attack capability and introduces threats to models, data and agent permissions; protection requires least privilege, provenance, testing, monitoring and human oversight. Future professionals must treat cryptography as part of a wider socio-technical system involving privacy, regulation, supply chains, safety and long-term maintainability.

Key Terms

Post-quantum cryptography: Cryptography designed to resist attacks from classical and quantum computers.

Shor’s algorithm: A quantum algorithm that can efficiently solve integer factorisation and discrete logarithms in theory.

Grover’s algorithm: A quantum search algorithm that provides a quadratic speed-up for unstructured search.

Harvest now, decrypt later: Collecting encrypted data today for possible decryption when stronger technology becomes available.

Cryptographic agility: The capability to replace algorithms, keys and parameters with limited system disruption.

Key encapsulation mechanism: A public-key mechanism for establishing a shared secret.

Hybrid cryptography: A construction combining classical and post-quantum mechanisms.

Lightweight cryptography: Cryptography designed for constrained devices and environments.

Trusted execution environment: An isolated execution area intended to protect code and data during processing.

Remote attestation: Evidence allowing a party to verify the measured state of a computing environment.

Homomorphic encryption: Encryption supporting specified computation directly on ciphertexts.

Secure multiparty computation: Protocols allowing joint computation while protecting private inputs.

Privacy-enhancing technology: A technology that reduces unnecessary disclosure or supports privacy-preserving use.

Differential privacy: A formal privacy framework limiting the effect of one individual’s data on an output.

Federated learning: Collaborative model training in which data generally remains near its original source.

Zero-knowledge proof: A proof of a statement that reveals no underlying secret beyond validity.

Prompt injection: Manipulation of an AI system’s instructions through crafted input.

Model extraction: An attempt to reproduce a model through queries or access to model artifacts.

Trusted computing base: The hardware and software components that must be trusted for a security property to hold.

Nonce: A value intended for one-time or unique use in a cryptographic protocol.

Review Questions

Short Answer Questions

1. Why are RSA and elliptic-curve systems vulnerable to a sufficiently capable quantum computer?
2. What is cryptographic agility, and why is it important during PQC migration?
3. State two operational risks associated with lightweight cryptography.
4. What security purpose does remote attestation serve?
5. Distinguish pseudonymisation from anonymisation.

Long Answer Questions

1. Compare lattice-based, hash-based and code-based post-quantum cryptography, including practical trade-offs.
2. Explain a staged post-quantum migration programme for an organisation with long-lived confidential data.
3. Describe how homomorphic encryption and secure multiparty computation protect privacy, and identify their limitations.
4. Explain the architecture and limitations of confidential computing.
5. Discuss how AI changes both the threat landscape and the protection requirements for cybersecurity systems.

Analytical / Application Questions

1. A manufacturer operates battery-powered sensors that cannot be physically accessed after deployment. Design a cryptographic lifecycle covering provisioning, nonce management, key rotation, firmware updates and device revocation.
2. Two hospitals want to calculate a joint disease statistic without sharing patient records. Compare MPC, federated learning and homomorphic encryption for this purpose, considering privacy, accuracy, communication and operational complexity.
3. An organisation stores encrypted legal records for 20 years and uses classical public-key encryption today. Develop a prioritised migration plan and justify which systems should be addressed first.

Multiple Choice Questions with Answers

Question 1

Which current public-key problem is threatened by Shor’s algorithm?

A. Integer factorisation

B. Collision resistance of every hash function

C. Random nonce generation

D. Physical destruction of storage media

Correct Answer: A. Integer factorisation

Question 2

What is the main purpose of a key encapsulation mechanism?

A. Compressing a digital signature

B. Establishing a shared secret

C. Removing metadata from a file

D. Detecting malware in firmware

Correct Answer: B. Establishing a shared secret

Question 3

Which property allows an organisation to replace cryptographic algorithms with limited redesign?

A. Data deduplication

B. Cryptographic agility

C. Traffic shaping

D. Air gapping

Correct Answer: B. Cryptographic agility

Question 4

Why are hybrid post-quantum deployments used during migration?

A. They eliminate the need for authentication

B. They combine classical and post-quantum mechanisms during transition

C. They guarantee that all implementations are side-channel free

D. They reduce every public key to one byte

Correct Answer: B. They combine classical and post-quantum mechanisms during transition

Question 5

Which technology protects data while it is being processed inside an isolated hardware-backed environment?

A. Confidential computing

B. Tokenisation alone

C. Password hashing

D. Packet fragmentation

Correct Answer: A. Confidential computing

Question 6

What does homomorphic encryption permit?

A. Computation on ciphertexts with a decryptable corresponding result

B. Recovery of plaintext without a key

C. Automatic removal of all statistical bias

D. Guaranteed prevention of denial-of-service attacks

Correct Answer: A. Computation on ciphertexts with a decryptable corresponding result

Question 7

In secure multiparty computation, the intended goal is to:

A. Publish every participant’s input

B. Compute a function while limiting disclosure of private inputs

C. Replace all network authentication

D. Make outputs independent of inputs

Correct Answer: B. Compute a function while limiting disclosure of private inputs

Question 8

Which statement about differential privacy is most accurate?

A. It guarantees that no individual can ever be inferred

B. It limits the effect of one individual’s data under a defined privacy mechanism

C. It requires publishing raw data first

D. It is identical to encryption at rest

Correct Answer: B. It limits the effect of one individual’s data under a defined privacy mechanism

Question 9

Which is an example of an AI-specific security threat?

A. Prompt injection

B. Key rotation

C. Certificate renewal

D. Secure erasure

Correct Answer: A. Prompt injection

Question 10

Why is nonce reuse dangerous in authenticated encryption?

A. It can undermine confidentiality or authentication guarantees

B. It always increases the key length

C. It prevents all replay attacks automatically

D. It converts symmetric encryption into public-key encryption

Correct Answer: A. It can undermine confidentiality or authentication guarantees