Lesson 1

Privacy and Personal Data

What data AI collects, how it's used, and what you can do about it.

In 2023, Italy temporarily banned ChatGPT over data privacy concerns — specifically that OpenAI had no legal basis for collecting and processing European users' personal data under GDPR. The ban was lifted after OpenAI added data controls, but the episode illustrated something important: AI systems collect and process data at a scale that even their users rarely understand.

What AI Systems Collect

When you use an AI app, it typically collects:

Conversation data: Everything you type, including personal details you share incidentally
Usage patterns: When you use it, for how long, what you ask about
Device and location data: Your device type, IP address, sometimes approximate location
Account data: If you're logged in, your profile information and history

Why It Matters

This data can be used to train future models, sold to advertisers, exposed in data breaches, or accessed by governments. Privacy settings and terms of service determine what a company can do with your data — but most people never read them.

Practical Steps

Check whether an AI app has a privacy setting to opt out of training data collection. Don't share sensitive personal information (medical, financial, relationship details) in AI chats. Assume anything you type could be read by a human reviewer.

Back Take Quiz

Quiz 1

Privacy and Personal Data

4 questions — free, untracked, retake anytime.

did Italy temporarily ban ChatGPT in 2023?

✓ Correct — ✅ ✓ GDPR requires a legal basis for data processing. Italy ruled OpenAI lacked this — a landmark moment for AI data privacy regulation.

❌ ❌ The ban was about data rights — specifically that OpenAI lacked legal basis under GDPR for collecting and processing European users' personal data.

data does a typical AI chat app collect beyond your messages?

✓ Correct — ✅ ✓ AI apps typically collect conversation content, usage patterns, device data, IP address, and account information — often more than users realize.

❌ ❌ Beyond messages, AI apps typically collect usage patterns, device information, IP address, and account data.

is a practical step to protect your privacy when using AI apps?

✓ Correct — ✅ ✓ Checking privacy settings and avoiding sensitive personal details in AI chats are practical, immediate steps everyone can take.

❌ ❌ Practical steps: check for training data opt-out settings, and avoid sharing medical, financial, or sensitive personal details in AI conversations.

should you assume that anything you type in an AI chat could be read by a human?

✓ Correct — ✅ ✓ Human review of AI conversations is common practice for quality and safety. Assuming a human might read your chat is the prudent default.

❌ ❌ Human review of AI conversations is standard practice. Treating your AI chat like a private diary is a mistake.

Back to Lesson Start Lab 1

Lab 1

Data Audit

Build a privacy evaluation framework for AI apps.

Lab 1 — Data Audit

Develop a personal data privacy framework for AI use.

The AI opens with the Italy/ChatGPT case and asks what data you think most people don't realize they're sharing.
Build a practical checklist for evaluating an AI app's data practices before using it.
Address: how do you balance utility vs. privacy when the most useful apps often collect the most data?

Consider: training data opt-outs, sensitive info categories, and what terms of service actually say.

🔬 AI GuideLab 1

Back to Quiz Next: Lesson 2

Lesson 2

Scams and Social Engineering

How AI has transformed phishing, fraud, and manipulation at scale.

In 2024, a UK energy company's CEO wired €220,000 to a fraudster after receiving a phone call from what he believed was his parent company's CEO — the voice was a perfect AI clone. In the same year, a study found that AI-generated phishing emails had a 50% higher click-through rate than human-written ones. Social engineering — manipulating people rather than hacking systems — has been turbocharged by AI.

How AI Upgrades Social Engineering

Voice cloning: Clone someone's voice from a few minutes of public audio — podcasts, YouTube, voicemails
Personalized phishing: AI can scrape your LinkedIn, social media, and public data to write a phishing email that references real details about your life
Scale: What once required a skilled human con artist can now run automatically against millions of targets
Real-time deepfakes: Live video deepfakes can now be run in real-time in video calls

Defense Principles

Verify out-of-band: If someone calls or messages asking for money or access, hang up and call them back on a number you already have
Pre-arranged code words: Families and teams can establish a shared code word to verify identity in suspicious situations
Slow down urgency: Urgency is a manipulation technique. Any request that creates pressure to act immediately deserves extra scrutiny

Core Defense

Verification through a separate channel is the most durable defense — no matter how convincing the voice or video, calling back on a known number defeats it.

Back Take Quiz

Quiz 2

Scams and Social Engineering

4 questions — free, untracked, retake anytime.

was the AI voice clone attack on the UK energy CEO so effective?

✓ Correct — ✅ ✓ The attack exploited trusted voice recognition — the CEO had no reason to doubt a voice he recognized. AI voice cloning makes this attack scalable.

❌ ❌ The attack worked because the CEO trusted the voice he heard. AI voice cloning made it possible to synthesize a convincing replica of a known, trusted person.

is 'out-of-band verification' the most durable defense against voice/video social engineering?

✓ Correct — ✅ ✓ Out-of-band verification bypasses the attack entirely: no matter how convincing the deepfake, calling back on a known number defeats it.

❌ ❌ Out-of-band verification works because it bypasses the attack channel entirely. However convincing the impersonation, calling a known number defeats it.

does urgency in a request deserve extra scrutiny?

✓ Correct — ✅ ✓ Urgency is a manipulation technique. Scammers create artificial time pressure to bypass your critical thinking. Slowing down urgency is a defense.

❌ ❌ Urgency is deliberately engineered to prevent you from thinking carefully. Any request creating pressure to act immediately deserves extra scrutiny.

makes AI-generated phishing emails more dangerous than traditional ones?

✓ Correct — ✅ ✓ Personalized phishing using real details about your life is far more convincing than generic attacks — and AI can automate this at scale.

❌ ❌ AI scrapes public data (LinkedIn, social media) to write personalized phishing emails referencing real details about your life — dramatically more convincing than generic attacks.

Back to Lesson Start Lab 2

Lab 2

Social Engineering Defense

Design a defense protocol against AI-powered fraud.

Lab 2 — Social Engineering Defense

Build your personal and organizational defense protocol against AI-powered social engineering.

The AI opens with the UK CEO voice clone case and asks what organizational protocol would have prevented it.
Design a verification protocol for high-stakes requests (money transfers, account access, sensitive data).
Address: what habits would you build to apply this consistently without creating friction in normal work?

Consider: out-of-band verification, pre-arranged codes, urgency handling, and escalation thresholds.

🔬 AI GuideLab 2

Back to Quiz Next: Lesson 3

Lesson 3

Screen Time and Wellbeing

Persuasive design, engagement optimization, and your relationship with AI tools.

Social media platforms discovered that optimizing for engagement — the metric they could measure — produced features that maximized anxiety, outrage, and compulsive checking rather than genuine wellbeing. AI tools face the same incentive: companies are rewarded for usage time, not for how you feel after using them. The metric that's easy to optimize is not the same as the outcome that's good for you.

Persuasive Design in AI

AI tools use several techniques to maximize engagement:

Variable reward: Unpredictable interesting outputs keep you checking and re-engaging
Completion pressure: Conversations feel unfinished; there's always one more question
Personalization: The system learns what keeps you engaged and serves more of it
Frictionless access: No waiting, no effort — immediate gratification lowers the cost of one more interaction

Engagement vs. Wellbeing

Research on social media found that passive consumption (scrolling) reduced wellbeing while active creation (making things) often improved it. The same principle likely applies to AI: using AI as a tool to create something may serve your wellbeing better than using it as entertainment to consume.

The Useful Question

Before opening an AI app: "What am I trying to accomplish?" If the answer is "I just want to see what it does," that's a signal to set a strict time limit.

Back Take Quiz

Quiz 3

Screen Time and Wellbeing

4 questions — free, untracked, retake anytime.

do AI companies optimize for engagement rather than user wellbeing?

✓ Correct — ✅ ✓ Engagement is measurable and directly tied to revenue. Wellbeing is harder to measure and not directly monetized — so it gets underweighted in product decisions.

❌ ❌ Engagement is measurable and commercially rewarded. Wellbeing is harder to measure and not directly monetized — so the incentive is to optimize engagement even when it diverges from wellbeing.

is 'variable reward' as a persuasive design technique?

✓ Correct — ✅ ✓ Variable reward (like a slot machine) creates compulsive engagement — you keep going because the next output might be the interesting one.

❌ ❌ Variable reward: unpredictable outputs create compulsive engagement. Like a slot machine, you keep engaging because the next response might be the interesting one.

on social media research, which type of AI use is more likely to support wellbeing?

✓ Correct — ✅ ✓ Research suggests active creation supports wellbeing better than passive consumption. Using AI to make something may serve you better than using it to be entertained.

❌ ❌ Research suggests active creation (making things) supports wellbeing better than passive consumption (scrolling/consuming). Using AI as a tool to create is more likely to serve you well.

is the 'useful question' to ask before opening an AI app?

✓ Correct — ✅ ✓ Having a clear purpose before opening an AI app is the single most effective way to prevent aimless, compulsive usage.

❌ ❌ 'What am I trying to accomplish?' is the key question. If you don't have a clear answer, that's a signal to either skip it or set a strict time limit.

Back to Lesson Start Lab 3

Lab 3

Wellbeing Audit

Audit your AI usage through a wellbeing lens.

Lab 3 — Wellbeing Audit

Audit your current AI usage through a wellbeing lens.

The AI opens with the engagement vs. wellbeing distinction and asks you to reflect on your current AI use patterns.
Identify one AI use pattern that serves you well (active/creative) and one that might not (passive/compulsive).
Design a specific change to the second pattern.

Be honest. The goal isn't to use AI less — it's to use it in ways that actually serve you.

🔬 AI GuideLab 3

Back to Quiz Next: Lesson 4

Lesson 4

When to Ask for Help

AI as a support tool — and its limits in crisis situations.

Crisis text lines and mental health organizations have studied how people use AI chatbots during emotional distress. The findings are mixed: AI can provide a low-barrier first point of contact that reduces stigma around asking for help. But AI chatbots have also failed in crisis situations — providing incorrect information about resources, engaging with suicidal ideation rather than redirecting to professional help, or simply not recognizing the severity of a situation.

What AI Can and Can't Do in Crisis

AI can:

Provide a low-barrier, judgment-free starting point for expressing difficult feelings
Offer general information about mental health resources
Be available at 3am when human support may not be immediately accessible

AI cannot:

Accurately assess crisis severity
Take real-world action — call for help, contact someone on your behalf
Replace clinical judgment or professional care
Maintain consistent relationship and accountability over time

Knowing When to Escalate

Clear Escalation Signals

Talk to a real person — a trusted adult, counselor, or crisis line — if you are feeling unsafe, having thoughts of self-harm, in a situation where physical safety is at risk, or experiencing distress that has lasted more than a few days. AI is not equipped to be your primary support in any of these situations.

If you or someone you know is in crisis, contact the 988 Suicide and Crisis Lifeline by calling or texting 988.

Back Take Quiz

Quiz 4

When to Ask for Help

4 questions — free, untracked, retake anytime.

is a genuine benefit of AI chatbots in mental health support?

✓ Correct — ✅ ✓ Low barrier and no judgment are genuine advantages — AI can be a first step that reduces stigma. But it's a starting point, not a destination.

❌ ❌ AI's genuine benefit in this context: low barrier to entry and no judgment, making it easier for some people to start expressing difficult feelings.

is AI not appropriate as a primary support in a mental health crisis?

✓ Correct — ✅ ✓ AI can't assess how severe a situation actually is, can't contact anyone for help, and can't provide the clinical judgment that crisis situations require.

❌ ❌ AI can't accurately assess crisis severity, take real-world action, or provide clinical judgment — all of which are essential in a genuine crisis.

situation clearly requires escalating to a real person immediately?

✓ Correct — ✅ ✓ Thoughts of self-harm or feeling unsafe require a real person — a trusted adult, counselor, or crisis line. Contact 988 (call or text) if needed.

❌ ❌ Thoughts of self-harm or feeling unsafe require a real person immediately. Contact 988 (call or text) or a trusted adult.

is the 988 Lifeline?

✓ Correct — ✅ ✓ 988 is the Suicide and Crisis Lifeline — free, confidential, available 24/7. Call or text 988 if you or someone you know is in crisis.

❌ ❌ 988 is the Suicide and Crisis Lifeline — a free service you can call or text 24/7 if you or someone you know is experiencing a mental health crisis.

Back to Lesson Start Lab 4

Lab 4

Support Mapping

Build a tiered support map from AI to professional help.

Lab 4 — Support Mapping

Map your support network and identify clear escalation thresholds.

The AI opens with the AI-vs-human support question and asks you to think about different levels of difficulty.
Build a tiered support map: AI starting point → trusted person → professional → crisis resource.
Address: what would make you hesitate to reach out to a real person, and how could you lower that barrier?

Note: if you're currently experiencing distress, please reach out to a trusted person or text/call 988.

🔬 AI GuideLab 4

Back to Quiz Next: Lesson 5

Lesson 5

AI and Your Data

Terms of service, data brokers, and the long-term implications of your AI footprint.

When Samsung engineers pasted proprietary source code into ChatGPT to ask for help debugging, the code was potentially incorporated into OpenAI's training data. Samsung subsequently banned the use of generative AI tools for work. The incident illustrated a principle most users haven't internalized: in many AI systems, what you type doesn't stay private — it may become training data, be reviewed by humans, or be retained indefinitely.

What Happens to Your Data

Different AI services have different data practices — but common patterns include:

Training data use: Your conversations may be used to train future model versions unless you opt out
Human review: Samples of conversations are often reviewed by human contractors for quality and safety
Data retention: Conversation history may be retained for months or years
Third-party sharing: Terms of service may permit sharing with affiliated companies or partners

Your AI Footprint

Over time, your AI interactions create a data footprint — a record of your questions, concerns, interests, and habits. This data has value to advertisers, insurers, employers, and governments.

Practical Rules

Never paste proprietary, confidential, or sensitive information into AI chats. Use the data opt-out settings every major AI service offers. Treat your AI conversations like email — potentially readable by others.

Back Take Quiz

Quiz 5

AI and Your Data

4 questions — free, untracked, retake anytime.

was the key lesson from the Samsung source code incident?

✓ Correct — ✅ ✓ The Samsung case illustrated that AI chat is not a private channel — content may be used for training or reviewed by humans. Sensitive information doesn't belong there.

❌ ❌ The lesson: content pasted into AI chats may become training data or be reviewed. Sensitive, proprietary, or confidential information doesn't belong in AI chat.

might have an interest in your AI conversation data over time?

✓ Correct — ✅ ✓ Your AI footprint — questions, concerns, interests, habits — has value to advertisers targeting you, insurers pricing you, employers evaluating you, and governments monitoring you.

❌ ❌ AI conversation data has value to advertisers, insurers, employers, and governments — all of whom may be interested in your questions, concerns, and habits.

does 'opting out of training data use' mean for a typical AI service?

✓ Correct — ✅ ✓ Training data opt-out stops your conversations from being used to improve future models — but doesn't necessarily affect retention, review, or other data uses.

❌ ❌ Training opt-out prevents your data from being used for model training — but retention, human review, and other data uses may continue.

should you think about an AI chat in terms of privacy?

✓ Correct — ✅ ✓ Treat AI conversations like email — potentially reviewed by humans, retained indefinitely, and subject to data requests. Never share anything you wouldn't want others to read.

❌ ❌ Treat AI conversations like email — potentially readable by others and retainable indefinitely. Never share what you wouldn't want others to read.

Back to Lesson Start Lab 5

Lab 5

Data Footprint Audit

Build your personal AI data hygiene checklist.

Lab 5 — Data Footprint Audit

Assess your AI data footprint and develop minimization practices.

The AI opens with the Samsung case and asks: what's the most sensitive information you've ever put into an AI chat?
Identify what data categories you should never put in AI chats.
Build your personal AI data hygiene checklist.

Consider: categories of sensitive info, opt-out settings, and the difference between what's technically allowed and what's prudent.

🔬 AI GuideLab 5

Back to Quiz Next: Lesson 6

Lesson 6

Staying in Control

Autonomy, dependency, and building a healthy long-term relationship with AI.

Researchers studying heavy AI assistant users have observed a pattern they call "cognitive offloading" — users increasingly delegating decisions, problem-solving, and even social interactions to AI. For some tasks this is clearly beneficial (using a calculator instead of doing arithmetic in your head). For others, the gradual atrophy of skills once delegated to AI may leave users less capable over time, not more.

The Dependency Question

Dependence on AI tools exists on a spectrum:

Tool use: You could do this without AI, but AI makes it faster or better. (Writing assistance, research, calculation)
Augmentation: AI enables you to do something you genuinely couldn't otherwise. (Language you don't speak, code you haven't learned)
Dependency: You've stopped maintaining the capability yourself. If AI disappeared, you couldn't function in this area.

Staying in Control

The goal isn't to use AI less — it's to remain the author of your own decisions, skills, and relationships.

Periodically do things without AI that you normally do with it — to check whether you're maintaining the underlying skill
Be the one who makes final decisions — AI can inform, but you decide
Notice when AI is shaping your opinions rather than informing them

The Core Principle

AI is a tool you use. You are not a tool AI uses. Keep that distinction clear and you'll maintain healthy control.

Back Take Quiz

Quiz 6

Staying in Control

4 questions — free, untracked, retake anytime.

is 'cognitive offloading' in the context of AI use?

✓ Correct — ✅ ✓ Cognitive offloading: delegating cognitive tasks to AI. Beneficial for some tasks; potentially harmful when it leads to atrophy of skills you still need.

❌ ❌ Cognitive offloading: delegating decisions and problem-solving to AI. Beneficial for things like calculation; potentially harmful when it atrophies skills you still need.

distinguishes 'augmentation' from 'dependency' in AI use?

✓ Correct — ✅ ✓ Augmentation = AI expands what you can do. Dependency = you've stopped maintaining the capability yourself and couldn't function without AI.

❌ ❌ Augmentation: AI enables genuinely new capabilities. Dependency: you've stopped maintaining the underlying capability — AI disappearing would leave you unable to function in that area.

is a good way to check whether you're developing dependency on an AI tool?

✓ Correct — ✅ ✓ Doing the task without AI occasionally is the clearest check. If you find you can't, that's a signal of dependency worth addressing.

❌ ❌ The clearest check: try doing it without AI. If you can't, you've likely developed dependency and may want to rebuild that capability.

does 'remaining the author of your own decisions' mean when using AI?

✓ Correct — ✅ ✓ AI can inform your decisions with information and analysis — but you make the call. Deferring entirely to AI's judgment is a form of dependency.

❌ ❌ Remaining the author of your decisions means using AI as input — not as the decider. You get the information, you make the call.

Back to Lesson Start Lab 6

Lab 6

Autonomy Audit

Build your personal AI autonomy and control framework.

Lab 6 — Autonomy Audit

Assess your AI dependency patterns and build a healthy control framework.

The AI opens with the cognitive offloading concept and asks you to identify one area where you've delegated more to AI than you'd like.
Distinguish your AI uses as tool use, augmentation, or dependency.
Design a specific practice to maintain autonomy in one area that matters to you.

The goal isn't to use AI less — it's to remain the author of your own decisions, skills, and thinking.

🔬 AI GuideLab 6

Back to Quiz Module Test

Module 4 Test

6 questions covering all lessons. Free, untracked, retake anytime.

banned ChatGPT in 2023 because:

✓ Correct — ✅ ✓ GDPR requires legal basis for data processing — Italy ruled OpenAI lacked this, a landmark moment for AI data privacy.

❌ ❌ The ban was about data rights — OpenAI lacked legal basis under GDPR for processing European users' personal data.

most durable defense against AI voice clone attacks is:

✓ Correct — ✅ ✓ Out-of-band verification bypasses the attack entirely — however convincing the clone, calling a known number defeats it.

❌ ❌ Out-of-band verification: contact the person through a separate known channel. No matter how convincing the clone, this defeats it.

core problem with optimizing AI for engagement is:

✓ Correct — ✅ ✓ Engagement is what gets measured and rewarded — but it's not the same as what's good for users. Social media showed this pattern, and AI faces the same incentive.

❌ ❌ Engagement is easy to measure and commercially rewarded. Wellbeing is harder to measure. When they diverge, companies optimize for engagement.

is the key limitation of AI in mental health crisis situations?

✓ Correct — ✅ ✓ AI can't assess how severe a situation is, call for help, or provide clinical judgment — all essential in a real crisis.

❌ ❌ AI can't assess crisis severity, take real-world action, or substitute for clinical judgment. 988 (call or text) is available for crisis support.

Samsung source code incident showed that:

✓ Correct — ✅ ✓ AI chat is not a private channel. Content may be used for training or reviewed. Sensitive information doesn't belong there.

❌ ❌ The lesson: AI chat is not private. Content may become training data or be reviewed. Sensitive and proprietary information doesn't belong in AI conversations.

difference between tool use and dependency in AI is:

✓ Correct — ✅ ✓ Tool use = could do it without AI, AI just helps. Dependency = the capability has atrophied, you can't do it without AI anymore.

❌ ❌ Tool use: AI helps but you could do it yourself. Dependency: you've stopped maintaining the capability — AI disappearing would leave you unable to function in that area.

Back Academy Home