◇ OpenCyber Academy
Bachelor's · Chapter 5 · English

Chapter 5 — Security literacy

Open, free learning for anyone who wants to protect people and technology. No account is required to read this text or its worked answers.

Chapter 5 — Security literacy

Lesson map

Lesson 05 published. Map, example, EX013.

flowchart TD
  N0["Learning outcomes"]
  N1["5.1 The CIA triad"]
  N2["5.2 Risk vocabulary"]
  N3["5.3 Classic attacks and AI variants"]
  N4["5.4 Basic practices"]
  N5["Synthesis"]
  N0 --> N1 --> N2 --> N3 --> N4 --> N5
  N5 --> EX["Worked example"]
  EX --> Q["Exercise EX013"]

Logical drawing

flowchart LR
  N0["Learning outcomes"]
  N1["5.1 The CIA triad"]
  N2["5.2 Risk vocabulary"]
  N3["5.3 Classic attacks and AI variants"]
  N4["5.4 Basic practices"]
  N5["Synthesis"]
  N0 --> N1 --> N2 --> N3 --> N4 --> N5 --> Q["EX013"]

[Learning outcomes]
    |
    v
[5.1 The CIA triad]
    |
    v
[5.2 Risk vocabulary]
    |
    v
[5.3 Classic attacks and AI variants]
    |
    v
[5.4 Basic practices]
    |
    v
[Synthesis]
    |
    v
[Worked example] --> [Exercise EX013]

Learning outcomes

You will (1) explain confidentiality, integrity and availability; (2) define threat, vulnerability, risk and control; (3) describe common attacks (phishing, malware, credential theft) and their AI-era variants; and (4) apply basic personal and organisational security practices.

5.1 The CIA triad

Confidentiality keeps information from unauthorised people; integrity keeps it accurate and unaltered; availability keeps systems usable when needed. For an AI chatbot: confidentiality of patient data in conversations; integrity of answers and bookings; availability of the service.

For Ask Lumen, a confidentiality failure would be showing one patient another's appointment; an integrity failure would be a wrong answer about clinic hours or a booking made for the wrong person; an availability failure would be the chatbot being unusable during a busy morning. Each needs different controls, so naming the property at risk helps choose the right one.

5.2 Risk vocabulary

A threat is something that could cause harm (an attacker, an accident). A vulnerability is a weakness. Risk combines likelihood and impact. A control reduces risk. Precise vocabulary helps teams agree on priorities.

Applied to Ask Lumen: a threat is a patient trying to see another's appointments; the vulnerability is the booking service account that can act for any patient; the risk is that combination's likelihood times the harm of exposed health data; the control is a policy check tying each booking to the logged-in patient. Using the words consistently stops discussions from mixing problems with solutions.

5.3 Classic attacks and AI variants

Phishing tricks people into revealing credentials; AI can generate more convincing, personalised phishing and voice clones. Malware compromises devices; AI-generated code can contain vulnerabilities if accepted uncritically. Credential theft gives attackers access; credentials pasted into chatbots can leak. And AI systems introduce new attack types: prompt injection, data poisoning, model extraction and evasion.

The new attack types affect the same three properties: prompt injection threatens integrity of actions, extraction threatens confidentiality of the model, poisoning threatens integrity of training data.

5.4 Basic practices

Use multi-factor authentication, keep software updated, apply least privilege, back up important data, and report suspicious events. For AI tools: do not paste secrets or personal data into unapproved services, check outputs before acting, and follow organisational policy on approved tools.

These practices apply to AI work as much as to any other: an API key for the model provider is a credential that needs the same care as a password. At LCH, staff are told which AI tools are approved for patient-related work and that patient details must never be pasted into others. Checking AI output before acting on it is the habit that would have caught Rina's invented reference.

5.5 Worked example (synthetic LCH case)

A receptionist at LCH received a voicemail apparently from a clinic director asking to change a supplier's bank details urgently. Rina's security-literacy training session used this synthetic case to discuss voice cloning and the control that stops it: verifying payment changes through a known phone number and a second approver, regardless of how convincing the request sounds.

The control works whether the voice is real or cloned, which is why process beats detection here. LCH also told staff that urgency in a payment request is itself a reason to slow down and verify.

5.6 Errors and limits

AI-specific risks often reuse classic paths: a cloned voice is phishing by another medium, and a leaked API key is credential theft. Relying on spotting fakes fails as generation improves, while callback verification keeps working. Pasting sensitive data into public tools can expose it to the provider's retention and to anyone who later obtains that data, so LCH's policy names approved tools explicitly.

Exercises

Synthesis

Security literacy — CIA, risk vocabulary, common attacks and basic practices — is the base on which AI-specific security is built.

Continue studying — chapter 05 of 40

← Chapter 04 (draft) · Course chapter index · Chapter 06 (draft) →