AI accelerates attacks and defense — humans still make the final call.
Your family receives an urgent call in a voice they recognize.
"Mom, I was in an accident. Please send $5,000 now."
The voice, rhythm, and expressions sound exactly like you — but you are sitting safely at home.
If the voice is real, does that prove the caller is real?
A few public clips are all an attacker needs to build your digital twin.
AI reproduces voice, appearance, and mannerisms from public samples.
🎭
AI can reproduce voice, appearance, and mannerisms from public samples.
Short videos, voice messages, livestreams, and public interviews provide enough samples for AI to create a convincing digital impersonation in real time.
The attacker did not hack any device. They used content you chose to share publicly.
A single instruction spawns convincing phishing in every language.
Generative AI removes old language clues and multiplies attack volume.
✉️
AI generates fluent, personalized phishing messages in any language — in seconds.
One prompt generates dozens of persuasive phishing messages in perfect English, Spanish, Mandarin — any target language — each tailored to a different victim.
Grammar errors used to be a reliable red flag. That red flag is gone.
Information you paste into AI tools can be extracted by others.
What you tell an AI assistant may not stay private.
💬
Attackers design prompts that make a model reveal information from user inputs or training data.
Protected information may come from something a user entered into the tool or from the large dataset used to train the model. An attacker designs prompts intended to reveal it.
Do not place personal, private, or security information into an AI-driven tool.
False information planted today becomes an AI's answer tomorrow.
Attackers can plant false information that future models learn from.
🌐
Attackers publish or modify online content so false information enters a future training set.
A forum post six months ago claimed a school uses "SecureGuard Pro" with a download link. Today an AI assistant confidently repeats this claim — with the same malicious link.
The original forum post may be deleted. The false claim is now baked into the model.
AI pieces together a person's life from scattered public crumbs.
AI scans and organizes public information into actionable intelligence.
🔎
AI automates the collection and cross-referencing of public information about a target.
Social media posts, professional profiles, public photos, and forum activity are automatically collected and cross-referenced to build a detailed picture of a person's relationships, routines, and interests.
Technical implementation returns in Units 4 and 5.
AI writes offensive code as fast as it writes helpful code.
AI helps write malware, modify applications, and discover vulnerabilities.
💻
AI can help write malware, modify applications, and search codebases for vulnerabilities.
An attacker prompts AI to generate a script that scans a codebase for common vulnerabilities or creates obfuscated malware variants that evade signature detection.
Technical implementation returns in Units 4 and 5.
The voice and video are perfect. How do you verify?
Question
A relative receives a video call showing your face and voice asking for an emergency wire transfer. The voice, expressions, and mannerisms are all perfect.
What evidence would prove the caller is really you?
Sensitive information entered earlier appears in a later response.
Question
A school employee previously pasted a private teacher contact list into an AI assistant. An attacker later submits carefully designed prompts that cause the model to reveal information from that list.
Which AI-enhanced attack is the adversary using, and what information is at risk?
A false claim survives long after the original post disappears.
Question
An attacker publishes a false claim and malicious download link across public forums. The content later enters an AI model's training data, and the model recommends the malicious link to students as if it were trustworthy.
Which AI-enhanced attack caused the model to repeat this false information?
Public details become a detailed target profile.
Question
An attacker uses AI to scan the CEO's professional profile, public social-media posts, conference appearances, and forum activity. The tool connects these scattered details into one organized profile.
Which AI capability is the adversary using, and how can it strengthen a later attack?
A prompt becomes an automated vulnerability scanner.
Question
An attacker asks an AI coding tool to create a script that searches a company's public-facing application for documented vulnerabilities. The attacker then runs the generated scanner across many pages.
Which AI capability is the adversary using, and how does it increase the attacker's capability?
AI amplifies impersonation, phishing, extraction, poisoning, reconnaissance, and coding.
Clone voice, image, and mannerisms from public video and audio samples.
Generate fluent, personalized phishing in any target language.
Craft prompts that make a model reveal information from user inputs or training data.
Plant false online content that future models learn and repeat.
Scan and cross-reference public sources to build detailed target profiles.
Help write malware, modify applications, and locate vulnerabilities.
Six attack vectors. Four defenses. One question — what would you do?
An attacker can clone a voice. They can't clone a private memory.
Agree on a private word or phrase only trusted people know.
🗝️
A private word or phrase only trusted people know cannot be cloned.
"What's the name of Grandpa's first dog?" — An AI clone might know public facts about you, but not a secret shared only between you and your mother.
Set this up in person during a calm moment, not during an emergency call.
Even a perfect clone hits a wall when MFA asks for the next factor.
A cloned voice or stolen password is not enough when a second factor is required.
🔐
A second authentication factor can prevent a cloned voice from being enough to access an account.
Even if an attacker clones your voice to pass voice authentication, they still cannot produce the one-time code sent to your phone.
Enable MFA wherever available. A cloned voice alone cannot provide a separate possession or authentication factor.
Patient records, passwords, and private keys don't belong in a prompt.
Do not enter personal, private, or security data into AI-driven tools.
🙊
Do not enter personal, private, or security data into AI-driven tools.
A doctor pastes patient records into a chatbot to summarize them. A student pastes exam questions. An engineer pastes a company's internal API keys to debug code.
Some AI tools feed user input back into model training. Treat every prompt as potentially public.
AI speaks with equal confidence about facts and fabrications.
Fluent output can still be wrong or manipulated.
✅
Fluent output can still be wrong or manipulated.
An AI confidently claims a security product is "industry standard" and provides a download link. The claim originated in a sponsored forum post by the product's manufacturer.
Use reputable, stable, non-AI sources to confirm security claims before acting on them.
Design a family response to a convincing digital impersonation.
Question
A relative receives a video call showing your face and voice asking for an immediate secret payment.
Write a verification protocol for this situation.
Protect identity, accounts, inputs, and decisions.
Use a family or friend shared secret in high-risk situations
Call back using a trusted number instead of the incoming call
Enable MFA wherever available
A cloned voice is not the same as possessing a second factor
Do not enter personal, private, or security data into AI tools
Treat chatbot and virtual-assistant prompts as data disclosure
Check important claims against reputable, stable, non-AI sources
Fluent output can still be wrong or manipulated
A defender cannot manually inspect every digital action in a large network.
→
Millions daily
🧑💻
One Security Team
Human attention is valuable — and limited.
One analyst reading one log at a time will never keep up.
Advance the workflow from raw activity to an informed response.
👆
Click a stage to begin
Work through the pipeline one stage at a time.
AI scans thousands of firewall rules and flags the one that's too open.
AI recommends safer settings — a technician decides.
⚙️
AI can review firewall rules or access controls and recommend safer options.
A firewall rule allows access from any IP address. AI flags it as overly permissive and suggests restricting to known office ranges. A security technician reviews the context before applying the change.
A knowledgeable security technician should review each recommendation before it is implemented.
AI scans through code to find the weak spots a human might miss.
AI identifies vulnerabilities — a programmer decides how to fix them.
🧩
AI can identify application vulnerabilities and recommend mitigations.
AI scans an application and flags a SQL injection vulnerability in the login form. It suggests parameterized queries as a fix. A knowledgeable programmer reviews the finding and implements the appropriate solution.
AI can suggest fixes, but a programmer must understand whether the suggestion is correct for this specific code.
AI learns what normal looks like — and raises a flag when it sees something strange.
AI proposes rules to catch threats — an engineer decides what to deploy.
📏
AI can recommend rules for automated systems that monitor malicious activity.
AI notices that legitimate users rarely access the admin panel at 3 AM from a new device, and proposes a detection rule that triggers when this pattern occurs. A detection engineer reviews and tunes the rule before deployment.
Poorly tuned detection rules generate noise. AI proposes — an engineer decides.
An employee account signs in from two distant countries within five minutes.
Question
An employee account signs in from New York at 10:00 AM and from Lagos, Nigeria at 10:04 AM. The system detects an impossible-travel pattern. Company policy requires uncertain sign-in events to be paused and reviewed before an account is disabled.
Determine the most appropriate system response under this policy.
Every false alarm makes the next one easier to ignore.
Harmless activity is classified as malicious.
🚨
A false positive means the system flagged harmless activity as an attack.
An online shopper in a new city is blocked from checkout because the AI flags their unfamiliar IP address. They miss a limited-time purchase — but they are a real customer.
Too many false positives waste analyst time and can hide real threats in the noise.
The most dangerous attack is the one the system never flags.
Malicious activity is classified as harmless.
🙈
A false negative means the system missed a real attack.
A user downloads a keylogger that records every keystroke and sends them to an attacker. The AI classifies the keylogger's network traffic as a routine software update — and the attacker collects passwords for months without detection.
AI provides speed. Humans provide context and accountability.
An employee on a business trip logs in from a new city late at night.
Question
An employee on a business trip logs in from a new city at 11 PM for the first time. The AI flags the login as suspicious and locks the account.
What mistake is the AI making, and what is the real-world consequence?
An attacker slowly exfiltrates data over three weeks using stolen credentials.
Question
An attacker slowly exfiltrates data over three weeks using credentials stolen in a phishing attack. The AI classifies each transfer as normal backup traffic.
What mistake is the AI making, and what is the real-world consequence?
Use automation to find and respond faster — with human oversight.
Review millions of digital events
Separate likely malicious activity from harmless activity
Alert human cybersecurity personnel
Take a configured corrective action when a threat is detected
Technicians review configuration advice
Programmers review code fixes
Detection engineers review proposed rules
Complete it on time and submit it before the deadline.
Check what comes next and arrive ready.
Key takeaways from this lesson.
AI Amplifies Attacks
It increases credibility, speed, scale, reconnaissance, and technical capability.
AI Accelerates Defense
It reviews large data volumes, identifies likely threats, and supports faster response.
Humans Decide
People verify identity, protect data, review recommendations, and remain accountable.