STACKCURVE
ISC2 CISSP · 2024 outline · 2026 AI guidance

The CISSP Field Guide.

An original study guide covering all eight CISSP domains. Each domain gives you the exam's angle, the core concepts in tables, memory hooks, a 2026 AI update, and a self-check. Mark a domain done when you can teach it to someone else.

Exam weight by domain · tap a segment to jump
Security & Risk Management16% Asset Security10% Security Architecture & Engineering13% Communication & Network Security13% Identity & Access Management13% Security Assessment & Testing12% Security Operations13% Software Development Security10%
00

The exam and the mindset

100–150questions (adaptive, English exam)
3 hrsmaximum time
700 / 1000scaled passing score
5 yrspaid experience in 2+ domains (1-year waiver possible)

The test is adaptive. It gets harder when you answer correctly, and it can stop at any point after question 100 once it's confident you passed or failed. You can't skip a question or go back to one.

What changed in 2026

2026 update
  • The outline did not change. The April 15, 2024 CISSP outline and its domain weights are still current. The September 1, 2026 refresh was for the CC exam, and CCSP refreshed on August 1, 2026.
  • April 2, 2026: AI exam guidance. ISC2 published Exam Guidance for Artificial Intelligence. It adds AI as a lens on all eight domains instead of creating a ninth domain. AI shows up two ways: as a system you have to secure, and as a tool defenders use. Each domain below has a callout for its AI topics.
  • April 1, 2026: smaller experience-waiver list. Fewer certifications now waive one year of the experience requirement. Training Camp reports that CEH, CISA, CRISC and OSCP were removed. Check ISC2's current list before you apply.

AI vocabulary you need for the 2026 lens

TermWhat it meansTypical control
Prompt injectionInstructions hidden in user input or in retrieved content (indirect injection) that override the model's intended behaviorTreat model output as untrusted, give the model least privilege, filter input and output, keep a human in the loop for actions
Data poisoningTampered training or fine-tuning data that changes model behaviorData provenance, integrity checks, curated sources
Model inversion / extractionRebuilding training data or cloning the model through repeated queriesRate limits, output limits, monitoring for query abuse
Membership inferenceWorking out whether a specific record was in the training setDifferential privacy, data minimization
Model driftModel performance drifts away from its baseline over timeMonitor drift. Unexplained drift can be a sign of compromise.
HallucinationOutput that sounds confident and is wrong, including package names that don't existVerification, dependency allow-lists
XAIExplainable AI, meaning you can trace why the model made a decisionA requirement for auditability and accountability
NHINon-human identity: AI agents, bots and service accountsOwnership, least privilege, short-lived credentials, access reviews

For context only (not confirmed exam scope): NIST AI RMF (Govern, Map, Measure, Manage), ISO/IEC 42001 (AI management system), and the EU AI Act.

Think like the exam

Be the advisor, not the technician

Pick the answer a senior risk advisor would choose: policy and process before tools, and business goals before technical elegance.

People come first

Human life and safety always win. Choose evacuation before saving data, and fail-safe doors for people.

Read for the qualifier

Watch for FIRST, BEST, MOST and PRIMARY. Several answers can be true, but only one fits that word. "First" usually means assess, classify, or get management approval.

Management owns the risk

Senior management is ultimately accountable. Security advises, and the business decides.

D1

Security & Risk Management

16% · biggest domain
What the exam is really asking: Can you tie security to the business, measure risk, follow the law and ethics, and keep the company running when things go wrong?

Core principles

GoalOpposite (DAD)Example controls
ConfidentialityDisclosureEncryption, access control, classification
IntegrityAlterationHashing, digital signatures, change control
AvailabilityDestruction / denialRedundancy, backups, DDoS protection

The triad is extended by authenticity (it's genuine), non-repudiation (you can't deny you did it, usually proven with a digital signature), and privacy.

Governance and the policy stack

DocumentMandatory?What it does
PolicyYesHigh-level management intent: the "why" and "what"
StandardYesSpecific required technologies or methods (e.g., AES-256 for data at rest)
ProcedureYesStep-by-step instructions
BaselineYesThe minimum security configuration for a system
GuidelineNoRecommendations and best practice

Due diligence

Doing the research and planning: knowing your risks and what to do about them.

Due care

Actually doing it: carrying out and maintaining the controls a reasonable person would.

Risk management

Risk is a threat exploiting a vulnerability, weighted by likelihood and impact. An asset is anything of value. An exposure is a moment when an asset is open to loss.

SLE = Asset Value (AV) × Exposure Factor (EF)
ALE = SLE × Annualized Rate of Occurrence (ARO)
Safeguard value = ALE(before) − ALE(after) − annual cost of safeguard

Worked example: a $200,000 asset with an EF of 25% gives an SLE of $50,000. With an ARO of 0.2 (once every 5 years), the ALE is $10,000. A control that costs more than that each year doesn't pay for itself.

ResponseMeaning
AvoidStop doing the risky activity
Mitigate / reduceAdd controls to lower likelihood or impact
Transfer / shareInsurance or outsourcing. You can transfer the cost but never the accountability.
AcceptFormally sign off on the risk (management decision, documented)
Reject / ignoreNever an acceptable answer

Residual risk is what's left after controls. Total risk is the risk before controls. Management decides whether the residual risk falls within risk appetite.

Control types

By functionExample
DirectivePolicy, signage
DeterrentWarning banner, visible guard
PreventiveFirewall, lock, separation of duties
DetectiveIDS, audit logs, CCTV review
CorrectivePatching, antivirus quarantine
RecoveryBackups, DR site
CompensatingAn alternative control used when the primary one isn't feasible

By category, controls are administrative (managerial), technical (logical), or physical.

Frameworks to recognize

  • NIST CSF 2.0 (2024): Govern, Identify, Protect, Detect, Respond, Recover. Govern is the new function.
  • NIST RMF (SP 800-37): Prepare → Categorize → Select → Implement → Assess → Authorize → Monitor.
  • ISO/IEC 27001 is the certifiable management system (ISMS). 27002 is the control guidance.
  • COBIT covers IT governance. ITIL covers IT service management. SABSA is business-driven security architecture. Zachman is an enterprise architecture matrix.

Threat modeling

STRIDE threatViolates
SpoofingAuthentication
TamperingIntegrity
RepudiationNon-repudiation
Information disclosureConfidentiality
Denial of serviceAvailability
Elevation of privilegeAuthorization

Other methods: PASTA (a seven-stage, risk-centric process) and DREAD (scores threats by damage, reproducibility, exploitability, affected users, and discoverability).

Law, IP and privacy

ProtectionCoversDuration (U.S.)
CopyrightExpression of an idea (code, books)Author's life + 70 years
TrademarkBrand names, logosRenewable every 10 years
PatentNovel inventions20 years from filing
Trade secretConfidential business infoAs long as it stays secret (protected by NDAs)
  • GDPR: notify the supervisory authority within 72 hours of a breach. The controller decides why and how data is processed. The processor handles it for them.
  • HIPAA covers health data. SOX covers financial reporting. GLBA covers financial institutions. PCI DSS is a contractual industry standard, not a law.
  • Kinds of law: criminal (prison, beyond reasonable doubt), civil (damages, preponderance of evidence), and administrative/regulatory.

Business continuity

MTD ≥ RTO + WRT   ·   RPO = how much data you can afford to lose
  • BIA (business impact analysis) comes first. It finds critical processes and sets MTD, RTO and RPO.
  • BCP keeps the business running (strategic, organization-wide). DRP restores IT after a disaster (tactical, a subset of the BCP).
  • Senior management must approve and sponsor the BCP.

People and the supply chain

  • Personnel controls: separation of duties, job rotation, mandatory vacation (catches fraud), NDAs, background checks, formal offboarding.
  • Awareness changes behavior for everyone. Training builds job skills. Education builds deep, long-term understanding.
  • Supply chain: assess vendors, set SLAs, keep a right to audit, and ask for an SBOM.

ISC2 Code of Ethics: the canons, in priority order

  1. Protect society, the common good, necessary public trust and confidence, and the infrastructure.
  2. Act honorably, honestly, justly, responsibly, and legally.
  3. Provide diligent and competent service to principals.
  4. Advance and protect the profession.

When two canons conflict, the higher one wins.

Memory hooks
  • Due Diligence = Do Detect (research). Due Care = Do Correct (act).
  • Ethics canons in order: "Society, Honor, Principals, Profession".
  • Risk responses: "A MAT": Avoid, Mitigate, Accept, Transfer. Never Ignore.
2026 AI update Treat AI as a supply chain risk. Assess AI vendors the way you assess any third party, and add questions on training data provenance, how model updates are delivered and tested, and where your data goes. A model update is a change, so it goes through change management.
Saved in this browser only
D2

Asset Security

10%
What the exam is really asking: Do you know what data you have, who owns it, how sensitive it is, and how to protect it from creation to destruction?

Data roles

RoleResponsibility
Data ownerSenior business person who is accountable. Sets the classification and approves access.
Data custodianUsually IT. Carries out the owner's decisions: backups, controls, maintenance.
Data stewardBusiness-side data quality and proper use
Data controllerDecides why and how personal data is processed (GDPR)
Data processorProcesses data on the controller's behalf (often a vendor)
Data subjectThe person the data describes

Classification

Government (high → low)

Top Secret · Secret · Confidential · Unclassified

Commercial (typical)

Confidential/Proprietary · Private · Sensitive · Public

Classify according to the damage that disclosure would cause. Classification drives handling, labeling, storage and destruction rules.

Data lifecycle and states

Create → Store → Use → Share → Archive → Destroy

StateMain protection
At restFull-disk, database or file encryption, plus access control
In transitTLS, IPsec, VPN
In useMemory protection, trusted execution environments (confidential computing)

Removing data for good (NIST SP 800-88)

MethodWhat it doesNotes
ClearOverwrite through normal commandsStops simple recovery. Fine for reuse inside the organization.
PurgeLab-resistant removal: degaussing, crypto-erase, secure-erase commandsDegaussing does not work on SSDs or optical media
DestroyShred, pulverize, incinerate, disintegrateThe most secure option. Best for SSDs holding highly sensitive data.

Data remanence is the data left behind after deletion. Crypto-erase (destroying the encryption key) works well in the cloud and on SSDs.

Protecting data in practice

  • Scoping means picking the baseline controls that apply. Tailoring means adjusting them to fit your mission.
  • DLP stops data leaving. DRM controls use after it's shared. A CASB gives visibility and control over cloud apps.
  • De-identification methods: masking (hide parts), tokenization (swap the data for a token stored in a vault), pseudonymization (reversible with a key), and anonymization (irreversible).
  • Retention: keep data only as long as law and business need require. Keeping it longer adds liability.
Memory hooks
  • Owner = accountable, Custodian = custody. The custodian holds the data but doesn't decide.
  • "SSDs laugh at magnets." For SSDs, use crypto-erase or physical destruction.
2026 AI update Training datasets, pre-trained models and model weights are assets. Inventory them, give them an owner, and classify them. A training set full of customer records inherits the classification of those records, and stolen weights mean stolen intellectual property.
Saved in this browser only
D3

Security Architecture & Engineering

13% · most technical
What the exam is really asking: Can you build security in from the start, using sound design principles, the right models, cryptography, and physical protection?

Secure design principles

Least privilege · Defense in depth · Secure defaults · Fail securely · Separation of duties · Keep it simple · Zero trust ("never trust, always verify") · Trust but verify · Privacy by design · Shared responsibility

Security models

ModelProtectsKey rules
Bell-LaPadulaConfidentialityNo read up (simple security), no write down (* property). The strong * property limits read and write to your own level.
BibaIntegrityNo read down (simple integrity), no write up (* integrity). A subject can't invoke a higher-integrity subject.
Clark-WilsonIntegrityAccess triple: subject → program → object. Well-formed transactions (TPs), constrained data items (CDIs), separation of duties.
Brewer-NashConflicts of interestThe "Chinese Wall." Access changes dynamically based on what you've already accessed.
Graham-DenningRights managementEight rules for securely creating and deleting subjects and objects and granting rights
Take-GrantRights transferDirected graph with take, grant, create and remove rules
Harrison-Ruzzo-UllmanRights integrityAccess-matrix model. Asks whether a right can leak.

Evaluating products: Common Criteria (ISO/IEC 15408)

A Protection Profile states what the customer needs. A Security Target states what the vendor claims. The TOE (target of evaluation) is the product being tested. Assurance runs from EAL1 (functionally tested) to EAL7 (formally verified design and tested). EAL4 (methodically designed, tested, and reviewed) is the common commercial target.

System weak points

  • TOCTOU (time of check to time of use) is a race condition between checking a resource and using it.
  • Covert channels are hidden communication paths. A storage channel writes to a shared resource. A timing channel modulates resource use over time.
  • Aggregation combines low-sensitivity facts into something sensitive. Inference deduces hidden facts from what's visible.
  • Emanations are electromagnetic leakage. TEMPEST shielding counters them.
  • Maintenance hooks and backdoors: remove them before production.

Hardware roots of trust: a TPM (chip on the board that stores keys and supports secure and measured boot), an HSM (dedicated device for managing keys), and a TEE (isolated execution area on the processor). Protection rings: ring 0 is the kernel, ring 3 is user applications.

Cloud and virtualization

ModelYou manageProvider manages
IaaSOS, apps, data, identitiesHardware, network, hypervisor
PaaSApps, data, identitiesPlus OS and runtime
SaaSData, identities, configurationNearly everything else

You always own your data and its accountability. Type 1 hypervisors run on bare metal. Type 2 run on a host OS. VM escape (breaking out of a guest into the host) is the major virtualization risk. Also know ICS/SCADA (availability and safety come first), IoT, embedded systems, containers, and serverless.

Cryptography

Symmetric

One shared key. Fast, used for bulk data. AES (128-bit block, 128/192/256-bit keys), ChaCha20, 3DES (legacy), Twofish. DES (56-bit) is broken.

Keys needed = n(n−1)/2

Asymmetric

A public and private key pair. Slow, used for key exchange and signatures. RSA, ECC (same strength with smaller keys), Diffie-Hellman (key agreement only), ElGamal, DSA.

Keys needed = 2n

GoalWhich key
ConfidentialityEncrypt with the recipient's public key. The recipient decrypts with their private key.
Signature (integrity + non-repudiation)Hash the message, then sign the hash with the sender's private key. Anyone verifies with the sender's public key.
  • Hashing is one-way and fixed-length. Use SHA-2 or SHA-3. MD5 and SHA-1 are broken. Salting defeats rainbow tables.
  • Modes: ECB leaks patterns (avoid it). CBC chains blocks. CTR turns a block cipher into a stream. GCM adds authentication (AEAD).
  • PKI: the CA issues X.509 certificates, the RA verifies identity, and revocation is checked through a CRL or OCSP (OCSP stapling).
  • Attacks: brute force, birthday (hash collisions), known- and chosen-plaintext, side-channel (timing, power), and meet-in-the-middle (why 2DES adds almost nothing).
  • Post-quantum: NIST published FIPS 203 (ML-KEM), 204 (ML-DSA) and 205 (SLH-DSA) in 2024. Quantum computing mainly threatens RSA and ECC. "Harvest now, decrypt later" is the reason to migrate early.

Physical security

Fire class (U.S.)FuelSuppress with
AOrdinary combustiblesWater, foam
BFlammable liquidsCO₂, foam, dry chemical
CElectricalCO₂, clean agent (no water)
DCombustible metalsDry powder
KCooking oilsWet chemical
  • Sprinklers: wet pipe (always full of water), dry pipe (for freezing areas), pre-action (best choice for data centers, because it needs two triggers), and deluge (all heads open at once).
  • Fence height: 3–4 ft deters casual trespassers. 6–7 ft is hard to climb. 8 ft with barbed wire deters determined intruders.
  • Power problems: fault (a short loss), blackout (a long loss), sag (short low voltage), brownout (long low voltage), spike (short high voltage), surge (long high voltage).
  • CPTED (crime prevention through environmental design): natural surveillance, natural access control, territorial reinforcement.
Memory hooks
  • Bell protects secrets: "can't read up, can't write down." Biba does the opposite, and the I in Biba stands for Integrity.
  • For confidentiality, use their public key. For a signature, use my private key.
  • 100 users need 4,950 symmetric keys but only 200 asymmetric keys.
2026 AI update Prompt injection is an architecture problem, not just an input-validation bug. Design so the model's output is untrusted, the model's tools have least privilege, and risky actions need a human approval. Treat explainable AI (XAI) as a security requirement, because you can't audit or defend a decision you can't trace.
Saved in this browser only
D4

Communication & Network Security

13%
What the exam is really asking: Can you design and secure networks, and match each attack and control to the right layer?

OSI model: layers, attacks and controls

#LayerData unitExamplesTypical attacks
7ApplicationDataHTTP, DNS, SMTP, SNMP, WAFInjection, phishing payloads
6PresentationDataEncoding, compression, encryption (TLS is often placed here on the exam)Malformed-format exploits
5SessionDataRPC, NetBIOS, session setupSession hijacking
4TransportSegment / datagramTCP, UDPSYN flood, port scans
3NetworkPacketIP, ICMP, IPsec, routersIP spoofing, smurf, route poisoning
2Data LinkFrameEthernet, ARP, switches, 802.1XARP poisoning, MAC flooding, VLAN hopping
1PhysicalBitsCabling, hubs, radioWiretapping, jamming

The TCP/IP model has four layers: Application (OSI 5–7), Transport (4), Internet (3), and Network Access (1–2). The TCP handshake is SYN → SYN-ACK → ACK.

Ports to know

PortServicePortService
20/21FTP143IMAP
22SSH / SFTP161/162SNMP
23Telnet (cleartext)389 / 636LDAP / LDAPS
25SMTP443HTTPS
49TACACS+1812/1813RADIUS
53DNS3389RDP
80 / 88HTTP / Kerberos110POP3

Secure channels

  • IPsec: AH provides integrity and authentication but no encryption. ESP adds confidentiality. Transport mode protects the payload only (host to host). Tunnel mode wraps the whole packet (gateway to gateway VPN). IKE negotiates the security associations (SAs).
  • TLS 1.3 is the current standard. SSL and early TLS versions are deprecated.
  • Wireless: WEP is broken. WPA2 uses AES-CCMP. WPA3 uses SAE, which resists offline dictionary attacks. Enterprise Wi-Fi uses 802.1X/EAP, and EAP-TLS (certificates on both sides) is the strongest.
  • 802.1X roles: supplicant (the device), authenticator (the switch or access point), and authentication server (RADIUS).

Network architecture

Firewall typeWorks atNotes
Packet filterL3–4Stateless, fast, no context
Stateful inspectionL3–4Tracks connections
Application proxyL7Deep inspection, slower
NGFWL3–7Adds app awareness, IPS and user identity
WAFL7 (HTTP)Blocks web attacks like SQL injection and XSS
  • Segmentation: a screened subnet (DMZ) for public-facing servers, VLANs, and micro-segmentation for workload-to-workload traffic.
  • Zero trust (NIST SP 800-207): no implicit trust based on location. A policy decision point (PDP) decides, and a policy enforcement point (PEP) enforces, on every request.
  • SASE combines networking and security services in the cloud. SDN separates the control plane from the data plane. NAC checks a device's posture before it can join.
  • Converged protocols: iSCSI and FCoE (storage over IP and Ethernet), VoIP, and MPLS.
Memory hooks
  • OSI layers from 1 to 7: "Please Do Not Throw Sausage Pizza Away".
  • IPsec: AH = Authentication Header only. ESP = Encrypts Secret Payload.
2026 AI update Segment AI workloads. Put model training, inference endpoints and vector databases in their own zones. Apply micro-segmentation and zero trust so a compromised AI service can't reach the rest of the network, and watch the traffic going out from AI systems.
Saved in this browser only
D5

Identity & Access Management

13%
What the exam is really asking: Can you prove who someone is, give them exactly the access they need, and take it away on time?

The access chain

Identification (claim an identity) → Authentication (prove it) → Authorization (what you may do) → Accountability (logs tie actions to you).

FactorExamples
Something you knowPassword, PIN
Something you haveSmart card, hardware key, authenticator app
Something you areFingerprint, face, iris
Context signalsSomewhere you are (location), something you do (behavior)

Multi-factor means factors of different types. A password plus a PIN is still just one factor.

Biometric accuracy

MetricMeaning
FRR (Type I)False rejection: a legitimate user is turned away (annoying)
FAR (Type II)False acceptance: an impostor gets in (dangerous)
CER / EERThe point where FRR equals FAR. Lower is better when comparing devices.

Access control models

ModelWho decidesWhere you'll see it
DACThe owner, at their discretionFile shares, ACLs
MACThe system, by comparing labels to clearancesMilitary, government
RBACJob roleMost enterprises
Rule-basedGlobal rules applied to everyoneFirewalls
ABACAttributes of the user, resource and environmentZero trust, cloud policy
Risk-basedA live risk scoreAdaptive and conditional access

Federation and SSO

ProtocolPurposeNotes
SAML 2.0Authentication (federated SSO)XML assertions from the identity provider (IdP) to the service provider (SP). Common for enterprise web SSO.
OAuth 2.0Authorization (delegated access)Access tokens. "Let this app read my calendar."
OpenID ConnectAuthentication on top of OAuthID token (JWT). Modern consumer and mobile SSO.
KerberosNetwork SSOKDC = Authentication Service + Ticket Granting Service. Symmetric keys, tickets, needs time sync (about 5 minutes of allowed skew). The KDC is a single point of failure.

RADIUS

UDP. Encrypts only the password. Combines authentication and authorization. Common for network access.

TACACS+

TCP 49. Encrypts the entire payload. Separates the three AAA functions. Common for device administration.

Identity lifecycle

  • Provision → Review → Revoke. Remove access promptly at termination, ideally before or at the moment the person is told.
  • Privilege creep is access that piles up as people change roles. Regular access reviews fix it.
  • PAM (privileged access management) adds just-in-time (JIT) elevation, password vaulting, and session recording.
  • Passwords under NIST SP 800-63B: length beats complexity, check passwords against breach lists, and don't force periodic changes without a reason. FIDO2 / passkeys resist phishing.
Memory hooks
  • "OAuth authorizes, OIDC identifies."
  • Type II error (FAR) is the worse one for security, because an impostor gets in.
  • TACACS+ uses TCP and encrypts the Total packet.
2026 AI update Govern non-human identities (NHIs). AI agents and service accounts need a named human owner, least privilege, short-lived credentials, and regular access reviews, just like people. An AI agent that acts for a user should not hold more rights than that user.
Saved in this browser only
D6

Security Assessment & Testing

12%
What the exam is really asking: How do you prove your controls actually work, and how do you report that to management?

Assessment types

Vulnerability assessment

Finds and ranks weaknesses (scanning, CVSS scores). It doesn't exploit them.

Penetration test

Exploits weaknesses to prove impact. Needs written authorization and agreed rules of engagement before it starts.

Pen test phases: Planning → Discovery → Attack → Reporting. Knowledge levels: full knowledge (white box / known environment), partial (gray box), and zero knowledge (black box / unknown environment).

Testing software

MethodHow it works
SASTStatic analysis: reads source code without running it (white box, early in the lifecycle)
DASTDynamic analysis: tests the running app from the outside
IASTAgents inside the running app watch execution
FuzzingFeeds in malformed input. Mutation fuzzing changes real inputs. Generation fuzzing builds inputs from a specification.
Misuse-case testingTests how an attacker would abuse a feature
Interface testingTests APIs, user interfaces and physical interfaces
Synthetic vs RUMScripted transactions vs real user monitoring of live traffic
Regression testingConfirms a change didn't break something that used to work

Audits and SOC reports

ReportCoversAudience
SOC 1Controls over financial reportingAuditors (restricted)
SOC 2Trust Services Criteria: security, availability, processing integrity, confidentiality, privacyCustomers under NDA
SOC 3A summary of SOC 2The public

Type I checks whether controls are designed correctly at one point in time. Type II checks whether they operated effectively over a period (usually 6–12 months), and it's the one you want from a vendor. Audits can be internal, external, or third-party (an independent firm).

Measuring and reporting

  • KPIs look back at performance. KRIs look ahead to rising risk.
  • SCAP standardizes the formats: CVE (vulnerability IDs), CVSS (severity scores), CCE (configurations), CPE (platforms).
  • Log review, account-management reviews, and backup verification (test the restore, not just the backup).
  • Write reports for the audience. Executives get risk in business terms. Technical teams get the details to fix.
Memory hooks
  • SOC 1 = $ (financial), SOC 2 = security, SOC 3 = shareable.
  • Type I = Instant (a point in time). Type II = Interval (a period).
2026 AI update AI red teaming is different from penetration testing. It probes model behavior with prompt injection, jailbreaks, data-extraction attempts and harmful outputs, along with the usual infrastructure. Plan it as its own engagement with its own scope and success criteria.
Saved in this browser only
D7

Security Operations

13%
What the exam is really asking: Can you run security day to day: monitor, respond, investigate, and recover?

Incident response

StepWhat happens
1 DetectionSpot it through alerts, logs or user reports, then triage
2 ResponseActivate the IR team and start documenting
3 MitigationContain it to stop the spread
4 ReportingNotify management, and regulators or customers if required
5 RecoveryRestore systems to normal operation
6 RemediationFix the root cause
7 Lessons learnedReview and improve the process

NIST SP 800-61 condenses this to Preparation → Detection & Analysis → Containment, Eradication & Recovery → Post-Incident Activity. Rev. 3 (2025) maps incident response to the CSF 2.0 functions.

Investigations and evidence

  • Investigation types: administrative (internal), criminal, civil, and regulatory. Each has its own standard of proof.
  • Chain of custody records who handled evidence, when, and why. Without it, evidence may not be accepted in court.
  • Admissible evidence must be relevant, material and competent (legally obtained).
  • Evidence types: real (physical), documentary, and testimonial. The best evidence rule prefers originals. The parol evidence rule says a written contract is the whole agreement. Hearsay is generally inadmissible.
  • Order of volatility (collect first to last): CPU registers and cache → RAM → swap and page files → disk → remote logs → archives and backups.
  • Work on a forensic copy (bit-for-bit image, verified with a hash), never on the original.

Monitoring and ongoing operations

  • SIEM collects and correlates logs. SOAR automates the response. UEBA learns normal behavior and flags anomalies. Threat hunting assumes a breach and searches for it.
  • Change management: request → review → approve → test → implement → document. Emergency changes still get documented afterward.
  • Patch management: test before deploying, roll out in stages, and verify.
  • Operational controls: need to know, least privilege, separation of duties, job rotation, two-person control.

Backups, resilience and recovery sites

BackupBacks upArchive bitRestore needs
FullEverythingClearsThe last full backup
IncrementalChanges since the last backup of any kindClearsFull + every incremental since (fast backup, slow restore)
DifferentialChanges since the last fullLeaves setFull + the last differential
RAIDMethodSurvives
0StripingNo failures (performance only)
1MirroringOne disk
5Striping with parity (3+ disks)One disk
6Double parity (4+ disks)Two disks
10Mirrored sets, then stripedOne per mirror
SiteReady inCost
HotMinutes to hours (fully equipped, data current)Highest
WarmHours to days (equipment present, data restored on arrival)Medium
ColdWeeks (space and power only)Lowest
Mobile / cloud / reciprocalVariesReciprocal agreements are hard to enforce

DR tests, from least to most disruptive: checklist (read-through) → tabletop (structured walk-through) → simulation → parallel → full interruption.

Memory hooks
  • IR steps: "Dirty Rats Make Rotten Restaurants Really Lousy". Detect, Respond, Mitigate, Report, Recover, Remediate, Lessons.
  • Differential = Different from the full. Restore needs only two sets.
  • A full interruption test is the last resort, because it can cause a real outage.
2026 AI update Watch for model drift. Set a baseline for model behavior and alert on unexplained changes. Drift can be a sign of data poisoning or tampering, not just stale data. Add AI systems to logging, SIEM coverage and incident playbooks.
Saved in this browser only
D8

Software Development Security

10%
What the exam is really asking: Can you build security into every phase of development, and govern the code you buy, borrow and generate?

Development approaches

MethodShapeSecurity angle
WaterfallSequential phasesSecurity gates at each phase; changes are hard
SpiralIterations driven by risk analysisRisk is assessed every loop
Agile / ScrumShort sprintsSecurity stories and abuse cases in the backlog
DevSecOpsCI/CD pipeline"Shift left": automated SAST, SCA and secrets scanning on every build

Security belongs in requirements, from the very first phase. Fixing a flaw costs more the later you find it.

Maturity models

CMMI levelName
1Initial (chaotic, ad hoc)
2Managed (repeatable)
3Defined (standardized)
4Quantitatively managed (measured)
5Optimizing (continuous improvement)

Also know OWASP SAMM (prescriptive: what you should do) and BSIMM (descriptive: what other organizations actually do).

Common flaws and their fixes

FlawBest fix
SQL injectionParameterized queries (prepared statements), input validation
Cross-site scripting (XSS)Output encoding, input validation, Content Security Policy
CSRFAnti-CSRF tokens, SameSite cookies
Buffer overflowBounds checking, memory-safe languages, ASLR/DEP
Race condition / TOCTOULocking, atomic operations
Broken access controlServer-side authorization checks on every request
Insecure deserializationDon't deserialize untrusted data; check integrity first

Validate input on the server side, since anything the client checks can be bypassed.

Databases

  • ACID: Atomicity, Consistency, Isolation, Durability.
  • Polyinstantiation keeps different versions of a record at different classification levels to block inference.
  • Views limit what users see. Watch for aggregation and inference attacks.

Code you don't write yourself

  • SCA (software composition analysis) finds vulnerable open-source components. An SBOM lists every ingredient.
  • Code signing proves who wrote the code and that it hasn't been altered. Protect code repositories and pipeline secrets.
  • For third-party software and APIs: assess the vendor, put security requirements in the contract, and test what you receive.
Memory hooks
  • CMMI: "I Manage Data Quite Openly". Initial, Managed, Defined, Quantitatively managed, Optimizing.
  • Injection → parameterize. XSS → encode the output.
2026 AI update Treat AI-generated code as untrusted code. Watch for hallucinated dependencies (packages that don't exist, which attackers can register under that name) and insecure defaults. Run the same SAST, SCA and peer review on it as on human code, and pin dependencies to an approved list.
Saved in this browser only
∑

Numbers & formulas for the night before

ItemValueDomain
SLEAV × EFD1
ALESLE × AROD1
Safeguard valueALE before − ALE after − annual costD1
ContinuityMTD ≥ RTO + WRTD1
GDPR breach notice72 hoursD1
Patent / Copyright20 years / life + 70D1
Symmetric keysn(n−1)/2D3
Asymmetric keys2nD3
AES block / keys128-bit block; 128/192/256-bit keysD3
DES key56 bits (broken)D3
Common CriteriaEAL1 → EAL7D3
Fence heights3–4 ft · 6–7 ft · 8 ft + barbed wireD3
OSI layers7 (TCP/IP model: 4)D4
KerberosPort 88; ~5 min time skewD5
TACACS+ / RADIUSTCP 49 / UDP 1812–1813D5
RAID 5 minimum3 disksD7
CMMI5 levelsD8
⇄

Flashcards

Tap a card to flip it. Use the arrow keys to move and the space bar to flip.

Question