The CIA triad
Every security failure breaks one of three promises about data
Confidentiality, integrity and availability: who can read it, whether it is true, whether it is there.
Security can sound like a list of tools. It is really three promises about information. Confidentiality: only the people allowed can read it. It is broken when a customer database is copied and sold. Encryption and access control keep it. Integrity: the information cannot change without anyone noticing. It is broken when someone swaps the bank details on an invoice. Hashes and digital signatures keep it. Availability: it is there when it is needed. It is broken when ransomware locks a hospital’s records. Backups and redundancy keep it. Every attack and every control in this course is about one or more of these three.
Security properties · Going deeper
Beyond the triad: authenticity, non-repudiation and AAA
Three more properties show up in every exam: authenticity, non-repudiation, and the AAA of identity.
Three more properties show up in every exam: authenticity, non-repudiation, and the AAA of identity. Properties that extend the triad. Authenticity the data or message really comes from who it claims to. A signed software update proves its publisher. Non-repudiation the sender cannot later deny sending it. A digital signature made with a private key only they hold gives this; a shared password cannot. Privacy the right of a person to control how their personal data is collected and used. Confidentiality is a means to it, not the same thing. Safety for systems that act in the physical world, such as a pump or a car, the first promise is that no one is harmed. AAA: how a system enforces the triad for people. Accounting, also called auditing, is what lets an investigator answer “who did this, and when?”. Lessons 7 and 8 cover the first two, lesson 11 the third. The attacker’s view: DAD. Each property has an opposite. Disclosure breaks confidentiality, alteration breaks integrity, destruction (or denial) breaks availability. Some textbooks call this the DAD triad. Exam habit: when a question describes an incident, first ask which of C, I or A it broke. The answer usually follows.
Properties that extend the triad
- Authenticity the data or message really comes from who it claims to. A signed software update proves its publisher.
- Non-repudiation the sender cannot later deny sending it. A digital signature made with a private key only they hold gives this; a shared password cannot.
- Privacy the right of a person to control how their personal data is collected and used. Confidentiality is a means to it, not the same thing.
- Safety for systems that act in the physical world, such as a pump or a car, the first promise is that no one is harmed.
AAA: how a system enforces the triad for people
Accounting, also called auditing, is what lets an investigator answer “who did this, and when?”. Lessons 7 and 8 cover the first two, lesson 11 the third.
The attacker’s view: DAD
Each property has an opposite. Disclosure breaks confidentiality, alteration breaks integrity, destruction (or denial) breaks availability. Some textbooks call this the DAD triad.
Threats and vulnerabilities
A loss needs a threat, a weakness and something worth taking
A threat uses a vulnerability to harm an asset. Break any link and the risk disappears.
Every security incident has the same three parts. Here is one from 2024. A threat actor: a ransomware gang, looking for money. A vulnerability: a remote-access portal that asked only for a password. The gang had a stolen one, so it walked in. An asset: the claims and patient data of America’s largest health payments processor. The gang spent nine days inside before it struck. The impact: the data of 192.7 million people, and pharmacies and hospitals unable to process claims for weeks. Now notice: break any link and none of this happens. You rarely remove the threat. But a second factor on that portal would have closed the weakness.
Risk vocabulary · Going deeper
Asset, threat, vector, vulnerability, exploit, risk, control
A threat actor uses a threat vector and an exploit against a vulnerability in an asset.
Likelihood and impact
Rank risks by how likely they are and how much they would hurt
Risk combines likelihood and impact. A matrix ranks risks so the money goes to the worst first.
You cannot defend against everything equally, so you rank. Draw a grid. Across: how likely a loss is, from rare to almost certain. Up: how much it would hurt, from negligible to severe. Place each risk. Phishing that steals a password: almost certain, moderate harm. Ransomware: possible, but severe. A lost laptop: likely, but the disk is encrypted, so the harm is small. Multiply the two ratings and sort. Phishing and ransomware both score fifteen, deep in the red corner. That is where the next dollar goes. The rest is watched, and scored again as things change.
Qualitative and quantitative risk · Going deeper · 1 of 2
Put a price on it: SLE, ARO and ALE
SLE = asset value × exposure factor. ALE = SLE × ARO. A control is worth it if it cuts ALE by more than it costs.
Likelihood and impact · Going deeper · 2 of 2
An average breach costs $4.44 million and lasts 241 days
IBM 2025: $4.44 million average, $10.22 million in the US, 241 days from intrusion to containment.
Impact is easiest to argue in money. In 2025, the average data breach worldwide cost four point four four million dollars. In the United States, more than ten million, the highest ever recorded. That covers investigation, notifying customers, lost business, fines and recovery. And the attackers had time. On average, two hundred and forty-one days passed from the first intrusion to containment: about eight months. Organisations that used automation widely in their security work found and contained breaches eighty days sooner, and paid almost two million dollars less. Speed is the lever.
Risk treatment
Reduce, transfer, avoid or accept: every risk gets one answer
Four risk treatments: reduce, transfer, avoid, accept. Acceptance is only valid below the risk appetite.
Once a risk is ranked, someone decides what to do about it. There are only four answers. Take one risk: customers can now upload files, and a file might carry malware. It scores fifteen. Reduce it: scan every upload, allow only images and PDFs, and store them apart. Six remains. Transfer it: buy insurance. The money is covered, but the downtime and the damage to reputation are still yours. Avoid it: do not build the feature at all. No risk, and no benefit. Or accept it: launch as it is. That is only allowed when what remains is below the organisation’s risk appetite, the dashed line. Fifteen is far above it, so here acceptance would be negligent.
The risk register · Going deeper
Owner, score, treatment, residual: one row per risk
A risk register records each risk’s owner, inherent risk, treatment and residual risk.
Control types
Some controls act before an attack, some during it, some after
Deterrent and preventive controls act before, detective during, corrective after.
A control is anything that lowers a risk. The first way to sort controls is by when they act. Here is an attack, in the middle of time. Before it, a deterrent control discourages the attempt: a warning banner, visible cameras. A preventive control stops it outright: a second factor, a firewall rule, a locked door. During it, a detective control notices: an intrusion alert, someone reading the logs. After it, a corrective control repairs the damage: restoring from backup, patching, rebuilding. Two more: directive controls tell people what to do, and compensating controls stand in when the usual control cannot be used.
Control selection · Going deeper · 1 of 4
Choosing controls that work together
Good control sets pair prevention with detection and correction, and cost less than the risk they reduce.
Good control sets pair prevention with detection and correction, and cost less than the risk they reduce. Why one type is never enough. Preventive controls fail silently: when a phishing email gets through the filter, nothing tells you. A detective control turns that silent failure into an alert; a corrective control limits how far it spreads. Design each important risk with at least one of each. Compensating controls in practice. A legacy machine cannot be patched isolate it on its own network segment and allow only the one connection it needs. A small team cannot separate duties have a manager review every payment log weekly. A payment system cannot meet a PCI DSS requirement as written the standard allows a documented compensating control that meets the intent and is validated each year. Selecting a control. Effective it measurably lowers likelihood or impact. Proportionate its yearly cost is lower than the reduction in expected loss. Usable people will not work around it; a control that slows everyone down gets bypassed. Testable you can check that it still works, which is what audits do (lesson 14). Exam pattern: “What type of control is X?” Identify when it acts. A camera is detective when recorded footage is reviewed, deterrent when its visibility is the point.
Why one type is never enough
Preventive controls fail silently: when a phishing email gets through the filter, nothing tells you. A detective control turns that silent failure into an alert; a corrective control limits how far it spreads. Design each important risk with at least one of each.
Compensating controls in practice
- A legacy machine cannot be patched isolate it on its own network segment and allow only the one connection it needs.
- A small team cannot separate duties have a manager review every payment log weekly.
- A payment system cannot meet a PCI DSS requirement as written the standard allows a documented compensating control that meets the intent and is validated each year.
Selecting a control
- Effective it measurably lowers likelihood or impact.
- Proportionate its yearly cost is lower than the reduction in expected loss.
- Usable people will not work around it; a control that slows everyone down gets bypassed.
- Testable you can check that it still works, which is what audits do (lesson 14).
Control types · Going deeper · 2 of 4
Each control has a category and a type, and exams ask for both
Categories: technical, managerial, operational, physical. Types: preventive, detective, corrective, deterrent.
The second way to sort controls is by what they are made of. Technical controls are enforced by systems: a firewall rule, an intrusion alert, a restore from backup. Managerial controls are decisions and documents: a change approval policy, a risk assessment, an incident response plan. Operational controls are things people do every day: training, an analyst reading logs, a guard on patrol. Physical controls work on the building: locked doors, cameras, fire suppression, fences. Put the two sorts together and every control has an address in this grid. A fence is physical and deterrent. An intrusion alert is technical and detective.
Control types · Going deeper · 3 of 4
No one person should be able to finish a risky action alone
Separation of duties splits a sensitive task across people, so fraud needs collusion.
Some of the most effective controls are about people, not technology. Follow a payment. A clerk creates a supplier. The clerk enters the invoice. A manager approves the payment. Treasury releases the money. Now suppose the clerk is dishonest. They can create a fake supplier and a fake invoice. But they cannot approve their own invoice, so the payment stops. For the fraud to work, two people must now agree to commit it, which is far rarer and far easier to spot. The same idea applies to systems: the person who writes code should not be the only one who can deploy it.
Security design principles · Going deeper · 4 of 4
The design principles behind every control
Least privilege, need to know, defence in depth, fail secure, secure by default, keep it simple.
Least privilege, need to know, defence in depth, fail secure, secure by default, keep it simple. Principles you will see in every lesson. Least privilege every person and program gets only the access its task needs, for only as long as it needs it (lesson 8). Need to know even with clearance, you see only the data your current work requires. Defence in depth independent layers of control, so one failure is not a breach (lesson 9). Separation of duties and dual control no single person can complete or conceal a sensitive action. Fail secure when a control breaks, it denies access rather than allowing it: a crashed firewall drops traffic. Fire exits are the exception: life safety means fail safe (open). Secure by default a system ships with the safe settings on; users must choose to weaken them. Keep it simple every extra component and setting is more to get wrong. Complexity is the enemy of security. Zero trust no request is trusted because of where it comes from; each is verified (lesson 9). Job rotation and mandatory holidays. Fraud often needs constant hiding: the person behind it never takes leave. Making staff take holidays, or rotating them between roles, hands the work to someone else who may notice. Professional ethics. ISC2 members commit to its Code of Ethics, whose four canons in order are: protect society, the common good and the infrastructure; act honourably, honestly, justly, responsibly and legally; provide diligent and competent service to principals; and advance and protect the profession. When canons conflict, the earlier one wins. The ISC2 CC exam tests the order of the four canons. Society comes first, the profession last.
Principles you will see in every lesson
- Least privilege every person and program gets only the access its task needs, for only as long as it needs it (lesson 8).
- Need to know even with clearance, you see only the data your current work requires.
- Defence in depth independent layers of control, so one failure is not a breach (lesson 9).
- Separation of duties and dual control no single person can complete or conceal a sensitive action.
- Fail secure when a control breaks, it denies access rather than allowing it: a crashed firewall drops traffic. Fire exits are the exception: life safety means fail safe (open).
- Secure by default a system ships with the safe settings on; users must choose to weaken them.
- Keep it simple every extra component and setting is more to get wrong. Complexity is the enemy of security.
- Zero trust no request is trusted because of where it comes from; each is verified (lesson 9).
Job rotation and mandatory holidays
Fraud often needs constant hiding: the person behind it never takes leave. Making staff take holidays, or rotating them between roles, hands the work to someone else who may notice.
Professional ethics
ISC2 members commit to its Code of Ethics, whose four canons in order are: protect society, the common good and the infrastructure; act honourably, honestly, justly, responsibly and legally; provide diligent and competent service to principals; and advance and protect the profession. When canons conflict, the earlier one wins.
Lesson complete
You finished “Security principles and risk”
Nicely done.