Emerging and Future IoT Security

Introduction

The Internet of Things is moving from isolated sensors and connected appliances toward intelligent, autonomous, safety-critical systems. Industrial robots, connected vehicles, smart energy networks, medical devices, agricultural systems and public infrastructure increasingly make decisions using data collected from distributed devices. This expansion creates security problems that cannot be solved only with passwords, firewalls and periodic software updates.

Future IoT security will depend on cooperation between artificial intelligence (AI), secure hardware, privacy-preserving computation, distributed trust, advanced communications and regulation. At the same time, attackers are also using AI, automation and large-scale computing. Security professionals must therefore understand both the opportunities and the limitations of emerging technologies.

Learning Objectives

After completing this chapter, students should be able to:

1. Explain AI-enabled defensive security and AI-driven attacks against IoT.
2. Describe federated learning and its privacy and security challenges.
3. Assess the security implications of autonomous IoT systems.
4. Explain how 5G and future 6G networks influence IoT security.
5. Compare blockchain, confidential computing and post-quantum cryptography.
6. Identify important regulatory, ethical and governance requirements.
7. Recommend future skills and security practices for IoT professionals.

1. Why IoT Security Is Changing

Traditional information technology commonly protects servers, user accounts and enterprise networks. IoT security must additionally protect constrained devices, physical processes, radio interfaces, sensors, actuators, gateways, cloud platforms and supply chains. A compromise may cause not only data loss but also unsafe physical action.

Several trends are driving change:

• Scale: Millions of heterogeneous devices may be deployed for years.
• Distribution: Devices operate outside controlled data centres.
• Convergence: Operational technology, IT, cloud computing and telecommunications are interconnected.
• Autonomy: Systems increasingly detect conditions and act without human approval.
• Adversarial adaptation: Attackers use automation, machine learning and supply-chain techniques.
• Long device lifetimes: A device installed today may remain in service beyond the life of its original vendor.

A useful security objective is to protect confidentiality, integrity, availability, authenticity, privacy and safety. For IoT, safety deserves separate emphasis: an authenticated command can still be dangerous if it is issued by a compromised but valid account or generated by a faulty model.

2. AI-Enabled IoT Defence

AI-enabled defence uses machine learning, statistical analysis, knowledge representation or automated reasoning to improve security decisions. Common applications include:

• anomaly detection in network traffic and sensor behaviour;
• malware and firmware classification;
• detection of unusual identities, locations or command sequences;
• vulnerability prioritisation;
• automated incident triage;
• predictive maintenance and detection of sensor faults;
• adaptive access control and security orchestration.

Detection models

A supervised model learns from labelled examples of benign and malicious activity. It may classify a device session as normal or suspicious. An unsupervised or semi-supervised model learns normal behaviour and identifies deviations. A one-class model is useful when attack examples are scarce, but unusual legitimate events can produce false alarms.

For binary classification, precision and recall are important:

Here, TP is true positives, FP is false positives and FN is false negatives. Precision measures the proportion of alerts that are correct; recall measures the proportion of actual attacks detected. These measures assume clearly defined positive and negative classes. In a highly imbalanced IoT environment, accuracy alone can be misleading because normal events may vastly outnumber attacks.

A security operations team may prefer high recall for safety-critical systems, while a small team with limited capacity may need higher precision to avoid alert fatigue. The decision must reflect consequence, response cost and the possibility of automated containment.

AI-assisted response

An AI system can recommend actions such as isolating a device, rotating credentials, restricting a protocol or requesting human approval. Fully automatic response is appropriate only when the action is reversible, the model is well validated and safety constraints are explicit. For example, blocking a suspicious internet connection may be safer than shutting down an industrial controller.

AI does not replace secure architecture. It cannot compensate for unsupported firmware, weak identity management or unsegmented networks. Models also require trustworthy data, monitoring, periodic retraining and protection against manipulation.

3. AI-Driven Attacks

Attackers can use AI to improve the speed, scale and plausibility of attacks. Examples include:

• generating convincing phishing and social-engineering messages;
• adapting malware to evade signatures;
• discovering exposed services and weak configurations at scale;
• producing polymorphic or automatically modified payloads;
• creating synthetic voice or video for fraud;
• poisoning training data;
• crafting adversarial inputs that cause a model to misclassify activity;
• using large language models to assist reconnaissance and code development.

An adversarial example is an input deliberately altered so that a machine-learning model produces an incorrect output. In IoT, this could involve changing sensor readings, radio signals or network features. Data poisoning occurs when malicious training data changes the model's future behaviour. Model inversion and membership inference can expose information about training data.

Defensive measures include signed and traceable datasets, separation of training and production environments, robust validation, adversarial testing, rate limits, human oversight and model monitoring. A model should fail safely: uncertainty should lead to conservative action or human review rather than unsafe control.

4. Federated Learning

Federated learning trains a shared model without collecting all raw data in one central repository. A coordinating server sends an initial model to selected devices or gateways. Each participant trains locally and sends an update rather than its raw records. The server aggregates updates and distributes a new model.

A common weighted aggregation rule is:

Here, w_t+1 is the global model after round t, w_t+1^(k) is the update or locally trained model from participant k, n_k is that participant's number of training examples, n is the total number of examples and K is the number of participants. This rule assumes that updates are comparable and that weighting by sample count is appropriate. In practice, local data may be non-identically distributed, devices may be unreliable and malicious participants may submit poisoned updates.

Federated learning reduces central data collection but does not guarantee privacy. A model update can leak information, and a malicious participant can corrupt the aggregate. Important protections include:

• Secure aggregation: the server sees only an aggregate, not each individual update.
• Differential privacy: controlled random noise limits information about one participant. Privacy improves as noise increases, but model utility may decline.
• Robust aggregation: rules such as coordinate-wise median or trimmed means reduce the effect of outliers, though no method works universally.
• Client authentication and attestation: only approved software contributes updates.
• Update clipping and anomaly screening: unusually large or inconsistent updates are investigated.

Federated learning is particularly useful when medical, industrial or household data cannot easily leave its source. It is less attractive when devices have insufficient processing power, unreliable connectivity or highly inconsistent data.

5. Autonomous IoT and Safety

An autonomous IoT system senses its environment, interprets information, selects actions and affects the physical world with limited human intervention. Examples include autonomous warehouses, connected vehicles, smart-grid control and precision agriculture.

Security must address the complete decision loop:

1. Sensing: Are sensors authentic, calibrated and resistant to spoofing?
2. Communication: Are messages confidential, fresh and authenticated?
3. Inference: Is the model reliable under unusual conditions?
4. Decision: Are policies and safety constraints enforced?
5. Actuation: Can commands be limited, logged and overridden?
6. Learning: Can feedback change behaviour without authorisation?

Autonomy introduces new failure modes. A compromised sensor can cause a correct algorithm to make an unsafe decision. A valid model can behave poorly outside its training distribution. An attacker may exploit interactions between individually secure components. Therefore, autonomous systems require defence in depth, independent safety controls, explainable operating limits, secure update mechanisms and a tested emergency stop or fallback mode.

The principle of least autonomy is useful: grant a system only the decision authority necessary for its task. High-impact actions should use thresholds, dual control or human confirmation where feasible.

6. 5G, 6G and IoT Security

5G supports enhanced mobile broadband, massive machine-type communication and ultra-reliable low-latency communication. Network slicing can create logically separated virtual networks for different services. Edge computing places processing closer to devices, reducing latency and bandwidth use.

These capabilities offer security benefits, including better isolation, local processing and stronger identity frameworks. They also enlarge the attack surface. Virtualised network functions, orchestration platforms, APIs, edge nodes and software-defined infrastructure become security-critical. A vulnerability in orchestration may affect many slices or tenants.

Important controls include:

• mutual authentication and strong device identity;
• secure boot and hardware-backed keys;
• encryption in transit and appropriate protection at the edge;
• isolation and policy enforcement between slices;
• continuous monitoring of APIs and orchestration systems;
• redundancy for critical connectivity;
• supply-chain assurance for network software;
• zero-trust access based on identity, device condition and context.

6G research is expected to combine AI-native networking, sensing and communication, integrated satellite and terrestrial systems, extremely distributed edge computing and new radio technologies. These are research directions rather than universally deployed specifications. Security design must therefore avoid assuming that an emerging capability is automatically mature or standardised.

7. Blockchain and Distributed Trust

A blockchain is a replicated, append-oriented ledger in which transactions are ordered through a consensus mechanism. In IoT, it may support device registration, provenance records, access agreements or audit trails. Smart contracts can automate rules such as authorising data exchange after defined conditions are met.

Blockchain can reduce dependence on a single administrative database, but it does not solve all trust problems. A ledger may prove that a statement was recorded, not that the original sensor was accurate. Private keys may be stolen, devices may be compromised before submission, and public ledgers may expose sensitive metadata. Consensus can also consume resources and introduce latency.

Blockchain is most appropriate when several organisations need a shared tamper-evident record and no single party is fully trusted. A conventional authenticated database is usually simpler for a single organisation. Good design stores hashes, identifiers or audit evidence on-chain while keeping personal or high-volume sensor data off-chain under appropriate access controls.

8. Confidential Computing

Encryption traditionally protects data at rest and in transit. Confidential computing seeks to protect data while it is being processed by using a hardware-supported trusted execution environment (TEE). A TEE isolates selected code and data from ordinary software, including some privileged system components.

Remote attestation allows a relying party to verify that an approved software measurement is running before releasing a key or data. The general trust sequence is: measure code, produce an attestation report, verify the report and policy, then release secrets.

Confidential computing can protect federated-learning aggregation, sensitive analytics and multi-organisation workloads. Limitations include side-channel attacks, reliance on hardware and firmware vendors, restricted debugging, availability concerns and the need to secure inputs and outputs. A TEE does not make malicious application code trustworthy; it protects the execution boundary under stated assumptions.

9. Post-Quantum Cryptography

A sufficiently capable quantum computer could threaten widely used public-key systems based on integer factorisation or discrete logarithms, including RSA and elliptic-curve cryptography. Symmetric cryptography is generally less affected, although larger security parameters may be selected.

Post-quantum cryptography (PQC) uses algorithms designed to resist known classical and quantum attacks. Major families include lattice-based, hash-based, code-based and other mathematical constructions. Standardisation and implementation guidance continue to evolve, so organisations should follow current national and international recommendations rather than selecting algorithms casually.

IoT migration is difficult because devices may have limited memory, processing capacity and update capability. A practical strategy is crypto-agility: design software and protocols so cryptographic algorithms and key sizes can be replaced without replacing the entire device. Organisations should inventory cryptographic use, protect long-lived confidential data, prefer authenticated updates and plan hybrid transitions where appropriate.

The “harvest now, decrypt later” threat is relevant when attackers record encrypted traffic today for future decryption. Sensitive data with a long confidentiality lifetime deserves priority in migration planning.

10. Regulation, Ethics and Governance

IoT regulation is increasingly concerned with secure design, vulnerability handling, privacy, transparency and accountability. Requirements differ by jurisdiction and sector, but common expectations include:

• security by design and by default;
• risk assessment before deployment;
• unique device identity and access control;
• secure update and vulnerability disclosure processes;
• support-period information;
• logging and incident reporting;
• protection of personal data;
• supply-chain and third-party risk management;
• evidence that safety and security controls are tested.

Privacy principles commonly include purpose limitation, data minimisation, storage limitation, accuracy and accountability. Consent alone is not a complete security strategy. A smart-building operator should ask whether occupancy data is necessary, who can access it, how long it is retained and whether individuals can exercise applicable rights.

Ethical issues arise when AI decisions affect employment, mobility, health or access to services. Organisations should define responsibility for automated decisions, test for unfair performance across groups, provide understandable explanations where required and retain meaningful human oversight. Security monitoring must also avoid becoming disproportionate surveillance.

Governance should assign ownership across the device manufacturer, integrator, network operator, cloud provider, data controller and user. Contracts should specify patch responsibilities, incident notification, evidence requirements and end-of-life handling.

11. Future IoT Security Skills

Future practitioners need a combination of technical, analytical and governance capabilities:

• embedded security, secure boot and hardware roots of trust;
• identity, certificates, key management and cryptographic agility;
• wireless, 5G/6G, edge and cloud security;
• AI security, data quality and model evaluation;
• privacy engineering and federated learning;
• threat modelling and safety analysis;
• vulnerability management and digital forensics;
• secure software development and supply-chain assurance;
• regulatory interpretation and risk communication;
• incident response and professional ethics.

Employability increasingly depends on the ability to translate technical risk into operational consequences. A security technician should be able to explain not only that a device is vulnerable, but also which process may fail, how quickly exploitation could occur, what evidence is needed and which mitigation is proportionate.

Applied Case: Illustrative Smart Water Network

Background

A regional water utility deploys connected pressure sensors, pump controllers and chemical-monitoring devices. Gateways send data to an edge platform, while a cloud service supports maintenance analytics. The utility plans to introduce federated learning so that several treatment sites can train a common fault-detection model without sharing raw operational data.

Problem

A compromised gateway submits manipulated model updates and causes the fault detector to ignore abnormal pump behaviour. At the same time, an attacker exploits an outdated public-key library in a long-lived controller. The 5G-connected maintenance architecture also contains a poorly isolated edge API.

Analysis

The incident combines data poisoning, software supply-chain risk, inadequate segmentation and cryptographic obsolescence. A blockchain ledger alone would not solve the problem because it could record false measurements. An AI detector might identify unusual updates, but automatic shutdown could interrupt water supply. Safety, availability and integrity must be considered together.

Approach

The utility adopts the following measures:

1. Devices receive unique identities, secure boot and signed firmware.
2. Gateways are remotely attested before joining the learning process.
3. Secure aggregation and update clipping reduce exposure to malicious contributions.
4. A robust aggregation method and human review are used for high-impact model changes.
5. Edge APIs are isolated, rate-limited and continuously monitored.
6. Controllers use a documented cryptographic-agility plan and a PQC migration assessment.
7. Independent safety controls retain local pump limits and manual override.
8. Contracts define patch support, incident reporting and end-of-life obligations.

Outcome

The utility reduces the likelihood that one compromised gateway can corrupt the shared model. It also improves detection of unsupported firmware and limits the consequences of an edge compromise. The solution does not eliminate risk: model drift, insider abuse and physical tampering remain possible and require continuing assessment.

Learning Points

Emerging technologies must be combined rather than treated as isolated remedies. Federated learning needs secure participants and robust aggregation; AI needs trustworthy data and human governance; 5G needs orchestration security; and cryptographic migration must begin before devices become impossible to update.

Challenges and Limitations

Future IoT security faces several persistent difficulties:

• constrained devices cannot run every security control;
• heterogeneous vendors use incompatible update and identity mechanisms;
• machine-learning performance changes as environments change;
• false alarms can overwhelm security teams;
• supply chains make responsibility difficult to assign;
• privacy protection may reduce analytic usefulness;
• autonomous actions can create safety consequences;
• emerging standards and regulations may differ across countries;
• security investment is often delayed because benefits are less visible than features.

A mature programme therefore uses risk-based prioritisation. It protects high-consequence assets first, documents assumptions, measures residual risk and revisits decisions as technology and threats change.

Best-Practice Checklist

1. Maintain an accurate inventory of devices, software, data flows and owners.
2. Use unique identities and least privilege rather than shared credentials.
3. Implement secure boot, signed updates and a realistic support plan.
4. Segment IoT, operational and administrative networks.
5. Monitor both cyber events and abnormal physical behaviour.
6. Validate AI data, models, updates and automated actions.
7. Apply privacy-preserving methods according to risk, not marketing claims.
8. Design for cryptographic agility and long device lifetimes.
9. Test incident response, fallback modes and manual override.
10. Record regulatory, ethical and supply-chain responsibilities.

Chapter Summary

Emerging IoT security is a multidisciplinary field shaped by AI, autonomy, advanced networks, distributed trust, protected execution and cryptographic change. AI can detect anomalies and support response, but it can also enable scalable attacks, poisoning and evasion. Federated learning limits central data collection but requires secure aggregation, participant authentication and robust defence against malicious updates.

Autonomous IoT requires protection of the complete sensing-to-actuation loop and must include safety constraints and human or independent control where appropriate. 5G and future 6G provide low latency, edge computing and virtualisation, while introducing risks in orchestration, APIs and software supply chains. Blockchain can provide shared auditability but cannot guarantee truthful sensor data. Confidential computing protects selected processing environments under explicit hardware and software assumptions. Post-quantum cryptography and crypto-agility are necessary for long-lived systems.

Effective future security combines technical controls with privacy, ethics, regulation, governance and professional skills. No emerging technology is a universal solution; security results from appropriate architecture, evidence-based risk management and continuous improvement.

Key Terms

AI-enabled defence: Use of artificial intelligence to support detection, analysis or response.

Adversarial example: An input deliberately designed to cause a machine-learning model to make an error.

Autonomous IoT: Connected systems that sense, decide and act with limited human intervention.

Blockchain: A replicated, append-oriented ledger maintained through a consensus mechanism.

Confidential computing: Protection of data during processing using a hardware-supported trusted execution environment.

Crypto-agility: The capability to replace cryptographic algorithms or parameters with limited redesign.

Data poisoning: Manipulation of training data or model updates to influence future model behaviour.

Federated learning: Collaborative model training in which raw data remains at participating locations.

Network slicing: Logical separation of virtual network resources for different services or tenants.

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

Precision: The proportion of positive classifications that are true positives.

Recall: The proportion of actual positive cases detected by a classifier.

Remote attestation: Verification that approved software is running in a specified environment.

Secure aggregation: A method that allows a server to obtain an aggregate of updates without viewing individual updates.

Security by design: Integrating security requirements throughout system conception, development and operation.

Supply-chain security: Protection against risks introduced by vendors, components, software and service providers.

Trusted execution environment: An isolated processing area intended to protect code and data from the surrounding system.

Zero trust: A security approach that continuously verifies users, devices and access context rather than trusting network location.

Review Questions

Short Answer Questions

1. Why is safety an additional concern in IoT security?
2. Distinguish precision from recall in AI-based intrusion detection.
3. What problem does secure aggregation address in federated learning?
4. State one benefit and one risk of network slicing.
5. What is meant by cryptographic agility?

Long Answer Questions

1. Explain how AI can support IoT defence and describe at least four ways attackers can exploit AI.
2. Describe the federated-learning process and analyse its main privacy and integrity risks.
3. Discuss the security architecture required for an autonomous IoT system.
4. Compare blockchain, confidential computing and conventional databases for IoT security applications.
5. Explain why IoT organisations should begin planning for post-quantum cryptography.

Analytical / Application Questions

1. A hospital uses connected infusion pumps and an AI anomaly detector. Recommend a response policy that balances recall, false alarms, patient safety and human oversight.
2. Design a security plan for a 5G smart-factory deployment containing edge servers, network slices and autonomous robots. Identify at least five controls and justify each.
3. An IoT vendor cannot update a ten-year-old sensor fleet and stores sensitive telemetry for future analysis. Analyse the technical, regulatory and business risks and propose a phased mitigation plan.

Multiple Choice Questions with Answers

Question 1

Which measure is most directly represented by TP ÷ (TP + FN)?

A. Precision

B. Recall

C. Specificity

D. Accuracy

Correct Answer: B. Recall

Question 2

What is the principal purpose of secure aggregation in federated learning?

A. To encrypt all sensor readings permanently

B. To allow the server to view each client model separately

C. To hide individual client updates while computing an aggregate

D. To eliminate the need for client authentication

Correct Answer: C. To hide individual client updates while computing an aggregate

Question 3

Which example is a data-poisoning attack?

A. Changing training records so that the deployed model learns incorrect behaviour

B. Encrypting a device disk with a strong key

C. Segmenting an IoT network into separate zones

D. Using remote attestation before releasing a key

Correct Answer: A. Changing training records so that the deployed model learns incorrect behaviour

Question 4

Why is an independent safety control important in an autonomous IoT system?

A. It guarantees that the AI model is always accurate

B. It provides protection if sensing, inference or control logic is compromised

C. It removes the need for software updates

D. It prevents all denial-of-service attacks

Correct Answer: B. It provides protection if sensing, inference or control logic is compromised

Question 5

Which 5G capability can create additional security risk through virtualised orchestration?

A. Network slicing

B. Manual paper logging

C. Offline storage only

D. Single-user local processing

Correct Answer: A. Network slicing

Question 6

What can a blockchain ledger establish most directly?

A. That every sensor measurement is physically accurate

B. That a recorded transaction has been made tamper-evident according to the ledger system

C. That private keys can never be stolen

D. That all data is automatically private

Correct Answer: B. That a recorded transaction has been made tamper-evident according to the ledger system

Question 7

Confidential computing primarily protects data while it is:

A. Being processed in a protected execution environment

B. Being printed on paper

C. Being deleted from a database

D. Being transmitted without authentication

Correct Answer: A. Being processed in a protected execution environment

Question 8

What is the purpose of crypto-agility?

A. To prevent any future algorithm change

B. To replace cryptographic mechanisms without redesigning the entire system

C. To remove encryption from constrained devices

D. To make passwords shorter

Correct Answer: B. To replace cryptographic mechanisms without redesigning the entire system

Question 9

Which issue is especially relevant to long-lived IoT devices when quantum computing advances?

A. Harvest now, decrypt later

B. Loss of all sensor calibration immediately

C. Automatic improvement of malware detection

D. Elimination of software supply chains

Correct Answer: A. Harvest now, decrypt later

Question 10

Which practice best supports security governance for a multi-vendor IoT deployment?

A. Assigning patch, incident-reporting and end-of-life responsibilities in contracts

B. Allowing every device to share one administrator password

C. Relying only on a public blockchain

D. Disabling all monitoring to protect privacy

Correct Answer: A. Assigning patch, incident-reporting and end-of-life responsibilities in contracts