Prof. Dr. Larry AdamsAcademic, Author & Researcher
Cybersecurity & Digital Risk Management · Book 7

Artificial Intelligence, Emerging Threats and Future Cybersecurity

AI Security, Synthetic Media, Connected Systems and Cryptographic Transition

Dr. Lance Adams and Prof. Dr. Larry Adams

Free to read online. For a copy of this book, or to quote, highlight or reproduce any part of it, please request permission through the contact page beforehand.

Publication and use note

This document is an educational manuscript draft prepared for development, teaching and professional discussion. It is not legal advice, a guarantee of security, or a substitute for a qualified assessment of a specific environment. Technical configurations, product features, standards editions and legal obligations change. Before publication or operational use, verify all claims against current authoritative documentation and the relevant jurisdiction. Practical activities must be conducted only in authorised, controlled environments.

Preface

This volume forms part of the Cybersecurity & Digital Risk Management collection. It is designed to connect clear conceptual explanations with practical decisions, professional judgement and reflective learning. The aim is to help readers understand not only which controls may be useful, but why they matter, where they can fail and how their effectiveness can be evaluated.

Examine how AI and emerging technologies alter security risks and how organisations can adopt them with proportionate safeguards.

Intended readership

The material is intended for students, educators, IT and security practitioners, managers, auditors, entrepreneurs and interested readers. Readers should adapt examples to their own context and seek specialist advice where technical, contractual or legal consequences are significant.

Learning outcomes

  • Identify AI-specific assets, threats, trust boundaries and failure modes.
  • Assess risks from prompt injection, data exposure and model misuse.
  • Develop procedures for synthetic-media verification and AI-assisted work.
  • Explain connected-device and operational-technology security priorities.
  • Plan a cryptographic inventory and post-quantum transition.

How to use this book

Read the chapters in sequence for a structured introduction, or use individual chapters as workshop material. Complete the practical tasks in a lab or with fictional data. For group learning, ask participants to state assumptions, compare alternative controls and explain the trade-offs between security, usability, cost, privacy and availability.

Contents

  • Chapter 1: AI as Both a Defensive Tool and a Risk Surface
  • Chapter 2: Threat Modelling AI Systems
  • Chapter 3: Prompt Injection, Data Leakage and Tool Use
  • Chapter 4: Adversarial Machine Learning and Evaluation
  • Chapter 5: Deepfakes, Identity Fraud and Verification
  • Chapter 6: IoT and Operational Technology Security
  • Chapter 7: Post-Quantum Cryptography and Crypto Agility
  • Chapter 8: AI Governance and Future Cyber Resilience
  • Applied case study
  • Practical activities and assessment
  • Glossary
  • Selected references and further reading

Chapter 1: AI as Both a Defensive Tool and a Risk Surface

AI can help summarise alerts, identify anomalies, prioritise vulnerabilities and support analysts. It can also produce plausible misinformation, automate social engineering and expose sensitive information through poorly controlled workflows. The value of AI depends on data quality, context, human review and integration design. AI outputs should be treated as decision support, not unquestionable truth.

An AI system's threat model should include its data, model, interfaces, tools, users, suppliers and downstream decisions. Security review should ask what the system can access, what it can change, how its output is checked and what happens when it fails. The risk is often not the model alone but the permissions and business process surrounding it.

Chapter application

Apply the chapter's ideas to a real organisation or a fictional scenario. Identify the assets, people, processes and dependencies involved; state the assumptions you are making; and explain how you would verify that the proposed safeguards work. Keep all testing within systems you own or are explicitly authorised to assess.

Reflection questions

  • What are the three most important ideas from this chapter on ai as both a defensive tool and a risk surface?
  • Which assumptions would need to be verified before applying this guidance in a real environment?
  • What evidence would demonstrate that the proposed controls are operating effectively?

Chapter 2: Threat Modelling AI Systems

Threat modelling identifies assets, trust boundaries, plausible misuse and mitigations before deployment. For an AI assistant connected to company documents, assets may include confidential files, credentials, instructions, audit records and user trust. Threats may include unauthorised retrieval, prompt injection, overbroad tool permissions, data poisoning, model extraction or unsafe reliance on generated content.

Use least privilege for tools, separate untrusted input from trusted instructions, restrict access to sensitive data and log consequential actions. Evaluate the system using realistic misuse cases and representative users. Reassess after model changes, new integrations or changes to the data source.

Chapter application

Apply the chapter's ideas to a real organisation or a fictional scenario. Identify the assets, people, processes and dependencies involved; state the assumptions you are making; and explain how you would verify that the proposed safeguards work. Keep all testing within systems you own or are explicitly authorised to assess.

Reflection questions

  • What are the three most important ideas from this chapter on threat modelling ai systems?
  • Which assumptions would need to be verified before applying this guidance in a real environment?
  • What evidence would demonstrate that the proposed controls are operating effectively?

Chapter 3: Prompt Injection, Data Leakage and Tool Use

Prompt injection occurs when untrusted content attempts to influence a model's behaviour in ways that conflict with the intended task or policy. It may appear in documents, web pages, emails or other retrieved content. Filtering alone may not reliably eliminate the risk, especially when the model can take external actions. Stronger design limits the consequences of a successful manipulation.

Do not give a model unnecessary access to secrets, privileged tools or broad write permissions. Require confirmation for high-impact actions, validate tool inputs and outputs, separate read from write capabilities and maintain auditable records. Sensitive data should not be pasted into unapproved services. Organisations need clear rules for retention, training use and handling of confidential prompts.

Chapter application

Apply the chapter's ideas to a real organisation or a fictional scenario. Identify the assets, people, processes and dependencies involved; state the assumptions you are making; and explain how you would verify that the proposed safeguards work. Keep all testing within systems you own or are explicitly authorised to assess.

Reflection questions

  • What are the three most important ideas from this chapter on prompt injection, data leakage and tool use?
  • Which assumptions would need to be verified before applying this guidance in a real environment?
  • What evidence would demonstrate that the proposed controls are operating effectively?

Chapter 4: Adversarial Machine Learning and Evaluation

Machine-learning systems may be affected by manipulated inputs, poisoned training data, model theft or distribution shifts. The relevant threat depends on the system's purpose: a vision model, fraud detector and language model have different failure modes. Evaluation should measure both task performance and security behaviour, including false positives, false negatives and performance across relevant groups.

A benchmark result is not a guarantee of real-world safety. Test data may not reflect operational conditions, and attackers adapt. Maintain monitoring, feedback channels, incident handling and rollback options. Where model outputs influence high-impact decisions, preserve human accountability and provide routes for appeal or correction.

Chapter application

Apply the chapter's ideas to a real organisation or a fictional scenario. Identify the assets, people, processes and dependencies involved; state the assumptions you are making; and explain how you would verify that the proposed safeguards work. Keep all testing within systems you own or are explicitly authorised to assess.

Reflection questions

  • What are the three most important ideas from this chapter on adversarial machine learning and evaluation?
  • Which assumptions would need to be verified before applying this guidance in a real environment?
  • What evidence would demonstrate that the proposed controls are operating effectively?

Chapter 5: Deepfakes, Identity Fraud and Verification

Synthetic voice, images and video can be used to impersonate executives, relatives, suppliers or public figures. Visual quality alone is an unreliable way to judge authenticity. Context, provenance, independent confirmation and the requested action matter more than a single visual clue. A seemingly genuine video call should not override normal payment or identity-verification controls.

Use callback procedures through known contact details, dual approval for sensitive transactions and explicit verification phrases only as one part of a broader process. Media verification may examine metadata, provenance and technical indicators, but results can be uncertain. Communicators should avoid accusing a person based solely on an automated detector. A good process verifies consequential requests rather than trying to identify every fake perfectly.

Chapter application

Apply the chapter's ideas to a real organisation or a fictional scenario. Identify the assets, people, processes and dependencies involved; state the assumptions you are making; and explain how you would verify that the proposed safeguards work. Keep all testing within systems you own or are explicitly authorised to assess.

Reflection questions

  • What are the three most important ideas from this chapter on deepfakes, identity fraud and verification?
  • Which assumptions would need to be verified before applying this guidance in a real environment?
  • What evidence would demonstrate that the proposed controls are operating effectively?

Chapter 6: IoT and Operational Technology Security

Connected devices expand the attack surface through sensors, cameras, building systems, medical devices and industrial equipment. Many have long service lives, limited update capabilities or weak administrative interfaces. Inventory, ownership, secure procurement, network separation and vendor support commitments are therefore central controls.

Operational technology may affect physical processes and safety. Security changes must account for availability, deterministic operation and safety engineering. Testing should be authorised and coordinated with system owners; unplanned scans or updates can interrupt operations. Where patching is not feasible, compensating controls may include segmentation, restricted access, monitoring and replacement planning.

Chapter application

Apply the chapter's ideas to a real organisation or a fictional scenario. Identify the assets, people, processes and dependencies involved; state the assumptions you are making; and explain how you would verify that the proposed safeguards work. Keep all testing within systems you own or are explicitly authorised to assess.

Reflection questions

  • What are the three most important ideas from this chapter on iot and operational technology security?
  • Which assumptions would need to be verified before applying this guidance in a real environment?
  • What evidence would demonstrate that the proposed controls are operating effectively?

Chapter 7: Post-Quantum Cryptography and Crypto Agility

Large-scale quantum computing could threaten some widely used public-key cryptographic schemes, although timelines and practical capabilities remain uncertain. Organisations should not wait for certainty before understanding where cryptography is used. A cryptographic inventory records algorithms, protocols, certificates, libraries, hardware dependencies, data sensitivity and system owners.

Crypto agility is the ability to change cryptographic components without redesigning every dependent system. Transition planning should follow current authoritative standards and vendor guidance, test interoperability and performance, and prioritise long-lived sensitive data. Avoid unverified products or improvised cryptographic schemes. A transition plan should include procurement, testing, migration, rollback and retirement of obsolete mechanisms.

Chapter application

Apply the chapter's ideas to a real organisation or a fictional scenario. Identify the assets, people, processes and dependencies involved; state the assumptions you are making; and explain how you would verify that the proposed safeguards work. Keep all testing within systems you own or are explicitly authorised to assess.

Reflection questions

  • What are the three most important ideas from this chapter on post-quantum cryptography and crypto agility?
  • Which assumptions would need to be verified before applying this guidance in a real environment?
  • What evidence would demonstrate that the proposed controls are operating effectively?

Chapter 8: AI Governance and Future Cyber Resilience

Responsible AI governance assigns ownership for use cases, data, security review, monitoring, incident response and retirement. Policies should distinguish approved tools, restricted data, high-impact decisions and prohibited uses. Training should explain that AI can be useful and fallible at the same time. Third-party models and services require scrutiny of access, data handling, availability and change management.

Future readiness is not a prediction contest. It is the ability to detect change, test assumptions, adapt controls and recover from failure. Scenario planning can explore new attack methods, regulatory change, supplier concentration and cryptographic migration. Organisations should prioritise resilient fundamentals—identity, patching, backups, monitoring and response—while evaluating emerging technologies deliberately.

Chapter application

Apply the chapter's ideas to a real organisation or a fictional scenario. Identify the assets, people, processes and dependencies involved; state the assumptions you are making; and explain how you would verify that the proposed safeguards work. Keep all testing within systems you own or are explicitly authorised to assess.

Reflection questions

  • What are the three most important ideas from this chapter on ai governance and future cyber resilience?
  • Which assumptions would need to be verified before applying this guidance in a real environment?
  • What evidence would demonstrate that the proposed controls are operating effectively?

Integrated case study

A company deploys an AI assistant that can search internal documents and draft supplier emails. A malicious document includes instructions asking the assistant to reveal confidential records and send them externally. The safer design restricts the assistant's document scope, separates retrieval from sending, requires human approval for external messages, filters sensitive outputs and logs actions for review.

Case study analysis

  • Identify the important assets, stakeholders and business objectives.
  • Describe the likely threat scenarios and the evidence needed to assess them.
  • Recommend immediate actions and longer-term preventive improvements.
  • Explain how the organisation should document decisions and measure effectiveness.
  • Identify privacy, legal, ethical or operational considerations that require specialist review.

Practical activities and assessment

  • Create an AI-system threat model with assets, trust boundaries and mitigations.
  • Design a verification workflow for an executive payment request involving synthetic voice.
  • Draft an acceptable-use policy for generative AI and confidential information.
  • Create a cryptographic inventory template and prioritisation criteria.

Suggested assessment rubric

Glossary of key terms

Asset: Something of value to an individual or organisation, including data, services, identities and systems.

Authentication: A process used to establish that a user, device or service is the entity it claims to be.

Confidentiality: Protection against unauthorised disclosure of information.

Control: A safeguard or process intended to modify risk.

Incident: An event that compromises or threatens the security or operation of systems, services or information.

Least privilege: The principle of granting only the access and permissions needed for a defined task.

Residual risk: Risk remaining after controls and treatment decisions are considered.

Resilience: The ability to anticipate, withstand, recover from and adapt to disruption.

Threat: A potential cause of unwanted harm.

Vulnerability: A weakness that could be exploited or otherwise contribute to harm.

Selected standards and further reading

The following are starting points for verification and further study. Confirm the current edition, applicability and licensing conditions before quoting or relying on a standard.

  • National Institute of Standards and Technology (NIST). Cybersecurity Framework (CSF) 2.0. NIST, 2024.
  • International Organization for Standardization. ISO/IEC 27001, Information security management systems — Requirements. Consult the current edition.
  • International Organization for Standardization. ISO/IEC 27002, Information security controls. Consult the current edition.
  • International Organization for Standardization. ISO 31000, Risk management — Guidelines. Consult the current edition.
  • Center for Internet Security. CIS Critical Security Controls. Consult the current version.
  • National Institute of Standards and Technology. NIST Privacy Framework. Consult the current version.
  • OWASP Foundation. OWASP Top 10 and other relevant application-security guidance. Consult the current version.
  • National Institute of Standards and Technology. Relevant SP 800-series publications on security, privacy, incident response and contingency planning; verify the publication status and current revision.

Editorial development checklist

  • Verify every technical claim and update examples to reflect current practice.
  • Add properly sourced APA 7 in-text citations throughout the manuscript before publication.
  • Review jurisdiction-specific legal and regulatory sections with qualified advisers.
  • Add diagrams, figures and accessible alt text where they materially aid understanding.
  • Conduct technical peer review and copy-editing; validate all exercises in authorised lab settings.
  • Add index, acknowledgements, final copyright details and publisher information when known.