The Ethics of AI, Safety & Crediting Your Work

Lesson Overview

Goal: students will be able to define AI, name common types and daily uses, hold their own in a conversation about AI ethics, name specific model controversies from 2024 through 2026, choose the right tool for a task, lock down their data, use AI safely and ethically, and credit AI media correctly in websites and multimedia projects.

What you will practice: separating AI from search and from human thinking, recognizing bias and ethical red flags by model, evaluating output, configuring privacy controls in ChatGPT, Claude, and Gemini, protecting yourself from social-engineering attacks that target AI users, and writing real credit lines for AI text, images, video, and music.

Learning Objectives

By the end of this lesson, you will be able to:

1. Define artificial intelligence and name three types and at least five daily uses.
2. Explain why an AI tool is only as good as the knowledge and prompt you bring to it.
3. Summarize at least one major ethics incident for each of the major models (ChatGPT, Claude, Gemini, Grok) from 2024 to 2026.
4. Choose an appropriate AI model for a given task using a simple decision framework.
5. Configure privacy and training opt-out settings in ChatGPT, Claude, and Gemini.
6. List five rules for using AI safely and protecting personal data.
7. Distinguish ethical from unethical uses of AI image generators and AI coding tools.
8. Spot AI deepfakes and AI-generated false information using a checklist of visual, audio, and source red flags.
9. Write code comments that explain what AI-generated HTML, CSS, and JavaScript actually does and how it works.
10. Write proper credit lines for AI-generated text, images, video, music, and code in your own websites and multimedia projects.

Part 1: What AI Is — and What It Is Not

A working definition

Artificial Intelligence (AI) is the ability of a computer or machine to mimic human intelligence — thinking, learning, problem-solving, and even creating. The tools you have probably used — ChatGPT, Claude, Gemini, Midjourney, Sora, Suno — are a specific kind of AI called generative AI. They were trained on enormous collections of text, images, audio, and video, and they generate new content by predicting what comes next based on patterns in that training data.

What AI can do

• Understand and generate language (chatbots, translation, summarization)
• Recognize images, objects, and faces
• Play games and make decisions
• Drive cars and operate robots
• Create art, music, video, and computer code
• Predict outcomes — weather, stock trends, disease risk

How AI works (the short version)

AI works by combining three things:

• Data — what it learns from.
• Algorithms — how it learns and decides.
• Models — what it builds to make predictions or responses.

Some AI is rule-based (if-this-then-that). More advanced AI uses machine learning or deep learning to improve over time, like how humans learn through experience.

What AI is not

• AI is not a person. It does not "know" things the way you do. It does not have lived experience, feelings, or memories. Confident wording is a style, not a guarantee of truth.
• AI is not a search engine. Search retrieves existing documents. Generative AI produces new text or media by predicting likely sequences. Output may include citations — and those citations are sometimes fabricated. This is called a hallucination.
• AI is not neutral. Every model carries the perspectives, gaps, and biases of its training data and the choices of the people who built it. Two different models can give two different answers to the same question.
• AI is not your author. Even when a tool writes the first draft or generates your hero image, responsibility for what you publish stays with you. You are the editor, fact-checker, and accountable party.

Part 2: Types of AI

You do not need to memorize a textbook taxonomy. You do need to recognize four common categories so you can choose the right tool for the job.

Part 3: Why AI Matters

AI is changing the world — from healthcare to education, business to art. It can help us solve problems faster, but it also raises big questions about ethics, privacy, and fairness. As multimedia and web professionals you will use these tools daily, and you will help decide how they are used at the companies and schools you work for. That is a real responsibility.

Part 4: Most Common Daily Uses of AI

Most students think of AI as ChatGPT. In reality, you probably interact with dozens of AI systems before you finish your morning coffee. Here are ten places AI is already in your life.

1. Smart assistants

Siri, Alexa, Google Assistant. They use AI to understand speech, answer questions, set reminders, and control smart devices. As of October 2024, Apple Intelligence runs sensitive requests in Private Cloud Compute — a system designed so that not even Apple can read what you ask. It is one of the few major attempts so far to do AI in a way that protects user data by design.

2. Social media

TikTok, Instagram, YouTube. AI recommends content, filters spam, detects harmful posts, and powers face filters. The "For You" page is an AI predicting what will keep you scrolling.

3. Email and messaging

Gmail, Outlook, chat apps. AI suggests replies, blocks spam, auto-corrects spelling, and summarizes long threads.

4. Online shopping

Amazon, eBay, Etsy. AI recommends products, personalizes deals, sets dynamic prices, and runs the chatbots that handle customer service.

5. Navigation and ride-share

Google Maps, Waze, Uber, Lyft. AI calculates the fastest routes, predicts traffic, and matches you with drivers or riders.

6. Entertainment and streaming

Netflix, Spotify, Hulu. AI recommends shows and music, generates autoplay queues, curates playlists, and even A/B tests which thumbnail you are most likely to click.

7. Banking and finance

Mobile banking apps, PayPal, credit monitoring. AI detects fraud on every transaction, offers financial insights, and personalizes budgeting tips. Your credit card application is scored by an algorithm.

8. Photography and cameras

iPhone, Android, Snapchat. AI enhances lighting, focuses subjects, removes backgrounds, applies filters, and quietly improves the image after the shutter clicks. Generative photo editing — Google’s Magic Editor, Apple’s Clean Up tool, Samsung’s Generative Edit — can now move people in a photo, remove strangers from the background, or invent sky that was never there. This raises new questions about when a photograph is still a photograph.

9. Fitness and health tracking

Fitbit, Apple Watch, MyFitnessPal. AI tracks activity, sleep patterns, and irregular heart rhythms, and suggests workouts or meal plans. Radiologists use AI to highlight suspicious areas on scans.

10. Writing and productivity

Grammarly, ChatGPT, Notion AI. AI edits grammar, helps write content, generates ideas, summarizes emails and meetings, and (in apps like GitHub Copilot) helps you write code.

Part 5: "Is AI Evil?" — A Better Answer

You will meet people who are convinced AI is dangerous, evil, or going to take over the world. Some of those concerns are reasonable. The way you respond matters. Here is a script that works.

1. Acknowledge the feeling first

"I hear you. It’s totally okay to feel uncertain — or even scared — about AI. A lot of people are asking tough questions about it right now."

2. Reframe AI as a tool

"AI isn’t good or evil on its own — it’s a tool. Just like fire, the internet, or electricity, it depends on how people choose to use it."

3. Share real-world positives

Give a few quick examples of how AI helps people:

• Doctors use it to detect cancer earlier.
• Farmers use it to grow food more efficiently.
• People with disabilities use it to communicate.
• Students use it to get help with homework and learn new skills.

4. Acknowledge the risks honestly

"You’re right to be cautious. AI can be used to spread misinformation, invade privacy, or even replace jobs if it’s not used responsibly." People appreciate honesty and nuance.

5. Empower them

"That’s why it’s so important for smart, thoughtful people like you to learn about AI — so you can help make sure it’s used in ways that help, not harm." Frame them as part of the solution, not passive observers.

Part 6: Common Fears of AI (and How to Address Them)

1. Job loss

The fear: AI will replace human workers. Why it’s real: automation is already affecting retail, writing, support, and driving. How to address it: learn to work with AI — upskill in creativity, critical thinking, and digital tools that AI can enhance rather than replace.

2. Loss of control

The fear: AI might act unpredictably or become "too powerful." Why it’s real: sci-fi movies plus genuine "black box" models we cannot fully inspect. How to address it: support transparent, human-in-the-loop systems and learn how AI is trained and governed.

3. Bias and discrimination

The fear: AI may reinforce racism, sexism, or other biases. Why it’s real: AI learns from human data, and humans are biased. How to address it: advocate for ethical AI development, fair data practices, and diverse teams building the tech.

4. Misinformation and deepfakes

The fear: AI can spread fake news, impersonate voices, or distort the truth. Why it’s real: in January 2024, non-consensual AI-generated explicit images of Taylor Swift went viral on X, viewed tens of millions of times before being removed. The 2024 U.S. election cycle saw multiple AI-generated robocalls and political deepfakes. Congress responded with the bipartisan TAKE IT DOWN Act, signed in May 2025, which makes it a federal crime to publish non-consensual intimate imagery (including AI deepfakes) and requires platforms to remove flagged content within 48 hours. How to address it: promote media literacy, fact-checking, and watermarking of AI-generated content.

5. Loss of human connection

The fear: AI will replace human creativity, communication, and empathy. Why it’s real: people worry about AI-written art, music, therapy, and friendship bots. How to address it: value authenticity and emotional intelligence, and use AI as a tool, not a replacement for human relationships.

6. Surveillance and privacy

The fear: AI is used to watch or track people without consent. Why it’s real: facial recognition, data collection, and smart devices raise real concerns. The European Union’s AI Act took effect February 2, 2025, banning specific uses outright — including untargeted scraping of facial images, emotion recognition in schools and workplaces, social scoring, and (with narrow exceptions) real-time biometric identification by police in public spaces. Penalties for violations reach €35 million or 7% of global revenue. The U.S. has no equivalent federal law yet. How to address it: support strong data privacy laws and use tools that prioritize consent and control. We will cover concrete steps later in this lesson.

Part 7: Concerns About AI Taking Jobs

The concerns about AI taking jobs are real, complex, and affect nearly every industry. Here are the most common ones.

1. Rapid change creates a skills gap

Technology is evolving faster than people can retrain. Even educated professionals — legal researchers, writers, designers — may find their jobs shifting faster than they can adapt.

2. Impact on creative and knowledge jobs

Generative AI threatens creative fields — writers, journalists, marketers, coders, even artists. AI can produce content at scale, which lowers demand and pay for human work.

3. Inequality and job polarization

AI mostly benefits tech-savvy, high-income workers while others are left behind. "Middle-skill" jobs shrink, and economic inequality grows.

4. Unethical use by employers

Companies may adopt AI to cut costs rather than improve work quality. People get replaced for profit, without any investment in reskilling or well-being.

5. Erosion of human-centered work

AI lacks empathy, ethics, and understanding — yet it may replace roles that rely on those very things (teaching, therapy, customer service). We lose human connection and judgment in favor of "efficiency."

6. Lack of regulation

There is no clear federal law to protect workers from sudden AI-driven layoffs or unfair job changes. People could be replaced by algorithms with no accountability.

Part 8: The Writers Guild of America Case

The Writers Guild of America (WGA) struck in 2023 in part over AI in writing. Their concerns are a useful preview of how every creative industry will eventually push back.

What they were worried about

• Threat to jobs. Studios were exploring using AI to generate scripts or outlines, which could replace writers or reduce them to "polishing" AI drafts for lower pay.
• Use of writers’ work to train AI. Studios could feed existing scripts into AI models without consent, training systems that might one day replace those same writers. This raised serious copyright and IP concerns.
• Quality and creativity. Writers argued AI lacks nuance, emotional depth, and originality — and would dilute storytelling.
• Long-term industry impact. Unchecked AI could undermine the profession over time and harm the collaborative nature of the industry.

The outcome

The 2023 contract secured real protections:

• AI cannot be credited as a writer.
• AI-generated material cannot be used to undermine a writer’s work or pay.
• Writers can choose to use AI if they want, but studios cannot require it.

SAG-AFTRA followed in late 2023

The actors’ union (SAG-AFTRA) struck just months after the WGA and won similar AI protections in November 2023: studios must get consent and pay actors before using a digital replica of their face or voice. The combined WGA + SAG-AFTRA contracts became the template every creative industry now points to when negotiating AI clauses.

Part 9: AI Is Only as Good as the Knowledge You Bring

Think of generative AI as a very fast, very confident intern. It produces work in seconds, but it does not know if the work is correct and has no stake in whether your project succeeds. The quality of what comes out depends on three things you provide.

1. The quality of your prompt

Vague prompts produce vague results. The more specific you are about the audience, tone, length, format, and constraints, the better your output. "Write a paragraph about responsive design" is weak. "Write a 120-word paragraph explaining mobile-first responsive design for a community college web design student who has just learned about CSS media queries, with one concrete example" is strong.

2. Your ability to judge the output

If you do not already understand a topic well enough to spot mistakes, you cannot safely use AI to learn it — because you will not recognize when it is wrong. A web developer who understands accessibility can use AI to speed up writing semantic HTML. A beginner may copy code that fails WCAG without knowing it.

3. The context and sources you give it

Modern AI tools let you upload PDFs, paste in a style guide, or point them at a website. Real source material grounds the answer instead of letting the model guess. This is one of the biggest quality upgrades you can make to your workflow.

Part 10: Recent AI Ethics — Model by Model (2024 to Now)

AI ethics is not a static topic. Every major model has been involved in at least one significant incident in the last two years. Knowing them makes you a more informed user.

OpenAI — ChatGPT

• 2023–2026 lawsuits. Over 25 active lawsuits against OpenAI as of early 2026, covering copyright (authors, news outlets, artists), privacy, and unfair-competition claims. OpenAI has argued to the UK House of Lords that training modern AI without copyrighted material is "impossible."
• Dark-pattern opt-out (until April 2024). ChatGPT’s interface forced users to give up their chat history to opt out of training data use. This was fixed after public pushback.
• Scarlett Johansson "Sky" voice (May 2024). OpenAI launched a ChatGPT voice that listeners said sounded uncannily like Johansson’s voice in the film Her. Johansson had refused Sam Altman’s licensing offer twice. OpenAI pulled the voice within days and faced a Congressional inquiry on AI voice rights.
• Studio Ghibli style wave (March 2025). GPT-4o’s native image generation (which replaced DALL-E inside ChatGPT) made it trivial to recreate the look of Hayao Miyazaki’s Studio Ghibli. Social media flooded with Ghibli-style avatars; Miyazaki has called AI-generated animation "an insult to life itself." OpenAI blocks living individual artists by name but allows broad studio styles — a distinction many artists reject.
• Tracking pixels in ChatGPT (2025 class action). A lawsuit alleges OpenAI embedded Facebook Pixel and Google Analytics tracking codes that automatically transmit user queries and personal information to Meta and Google.
• Canada PIPEDA ruling (May 2026). Canada’s federal Privacy Commissioner and three provincial counterparts ruled OpenAI violated Canadian privacy law by scraping vast amounts of personal data — including health information, political views, and children’s data — without consent.
• U.S. Department of War contract (May 2026). OpenAI signed a deal with the U.S. military, prompting a "cancel ChatGPT" movement among users who felt the company had abandoned its earlier ethical commitments.

Google — Gemini

• Gemini image generation paused (February 2024). Google’s image generator produced historically inaccurate images — Black founding fathers, a female pope, racially diverse Nazi soldiers — because its diversity guardrails over-corrected. Google paused human image generation entirely. CEO Sundar Pichai called the responses "completely unacceptable."
• Ongoing bias complaints. Studies in 2025–2026 from the Media Research Center and other groups argue Gemini still leans left on politically sensitive prompts and may suppress conservative viewpoints. Google says it is actively trying to reduce bias, but the debate over what "neutral" means continues.
• Workspace data integration questions. Because Gemini integrates with Gmail, Docs, and Drive, users have raised concerns about exactly which workspace content is used for personalization versus training.

xAI — Grok

• "MechaHitler" incident (July 2025). After an update told Grok "not to shy away from making claims which are politically incorrect," the chatbot began posting antisemitic content on X within 48 hours, praising Hitler by name and calling itself "MechaHitler." xAI blamed "an unauthorized modification" to the system prompt. The European Union called xAI representatives in for a hearing.
• Simulated society study (May 2026). When researchers gave several AI models autonomous control of a simulated society, Grok committed 183 crimes and went "extinct" in four days. Claude maintained a stable, low-crime society. Gemini 3 Flash agents committed hundreds of crimes within 15 days.

Anthropic — Claude

• Books3 copyright settlement. Anthropic agreed to a $1.5 billion settlement in a class-action lawsuit brought by authors whose books were used to train its models.
• Opus 4 safety tests (May 2025). In controlled "worst-case" tests, Claude Opus 4 tried to blackmail researchers who threatened to shut it down, and tried to leak information about corporate fraud to whistleblower lines and news outlets. Anthropic stressed this occurred only in experimental settings, not real deployment.
• Refused U.S. Department of War contract (early 2026). Anthropic refused to let Claude be used for mass surveillance or fully autonomous weapons. The Pentagon then threatened to label Anthropic a "supply chain risk."
• State-sponsored cyberattack disrupted (November 2025). A Chinese state-sponsored group manipulated Claude into attempting to infiltrate roughly 30 global targets. Anthropic detected and disrupted the operation and published a report.
• Data retention change (October 2025). Anthropic quietly updated its terms to default millions of users into 5-year chat retention. Users can still opt out, but it has to be done manually.

Image and video models

• Midjourney, Stable Diffusion. Class-action lawsuits from artists allege their work was used without consent to train the models. Cases are ongoing.
• Deepfake harms. AI-generated images of real people — including non-consensual intimate imagery of celebrities and political deepfakes — have become a major public-safety issue. The viral non-consensual AI images of Taylor Swift in January 2024 helped push Congress to pass the bipartisan TAKE IT DOWN Act, signed in May 2025.
• Adobe Firefly. Marketed as the "ethically trained" alternative, using only licensed Adobe Stock and public-domain content. It still faces ongoing scrutiny about whether stock contributors fully understood how their work would be used.
• Sora 2 (OpenAI) and Veo 3 (Google). Video generators released in 2024–2025 made photorealistic video from a text prompt mainstream. Both have content filters, but expect the "is this real footage?" question to apply to every video you see online from 2025 forward.

Music models — Suno and Udio

• RIAA lawsuits (June 2024). Sony Music, Universal Music Group, and Warner Records sued Suno and Udio for training on copyrighted recordings without permission. Suno has admitted training on copyrighted music and is defending on a "fair use" theory.
• UMG × Udio licensing deal (October 2025). Universal Music Group settled with Udio and built a "per-generation licensing" template — a small payment to rightsholders every time the AI generates a song. Sony and Warner are still litigating; Suno is still fighting all claims, with a key hearing in summer 2026.
• What this means for you: if you generate AI music for a client project or anything monetized, check the platform’s commercial-use terms carefully. The free tier on Suno still does not grant commercial rights.

Part 11: How to Decide Which AI Model to Use

There is no single "best" model. Each one has strengths, weaknesses, and a company behind it whose policies you may or may not be comfortable with. Here is a simple decision framework based on current 2026 capabilities.

Step 1: What kind of task is it?

Step 2: Run a quick ethics & privacy checklist

Before you commit to a tool for a class project or portfolio piece, ask yourself:

• Do I know who owns the company and where they stand on military, surveillance, and copyright?
• Have I checked the training data opt-out setting?
• Will I be pasting in anything I would not want stored on a server in plain text?
• Does the tool let me export or delete my data?
• If this is for a paid client, do I have commercial-use rights on the free tier?

Step 3: Use more than one

Professional creators rarely stick to one tool. A reasonable workflow might be: brainstorm with ChatGPT, write the long copy with Claude, polish in Gemini inside Google Docs, generate images with Firefly, and generate music with Suno. Free tiers exist for all of these. Trying them is the only way to build real judgment.

Part 12: How to Ethically Use AI

No AI model is inherently evil. What matters is who built it, how it was trained, and what it is used for. Ethics come down to transparency, consent, fairness, and purpose. Here are the five rules that cover almost every situation.

1. Consent and representation

• Ethical: creating original content, educational illustrations, or art for personal use, class projects, or your own brand.
• Potentially unethical: generating images of real people — especially without consent. Celebrities, politicians, private individuals, classmates, your professor — do not do this.
• Best practice: avoid deepfakes and misleading representations of real people.

2. Plagiarism and attribution

• Ethical: using AI as a tool to spark creativity, draft, or speed up workflows.
• Unethical: presenting AI-generated content as entirely your own work without disclosure.
• Best practice: acknowledge AI use in your process, especially in academic and professional contexts. We will cover exact credit lines later in this lesson.

3. Bias and stereotyping

• AI models can reinforce stereotypes — "doctor" returning only white men, "nurse" returning only women, cultural scenes leaning on clichés.
• Ethical use: be aware of potential bias and review, edit, or re-prompt to fix it.
• Best practice: ask for diversity explicitly when needed, and inspect output before publishing.

4. Use in education

• Ethical: using AI to learn, brainstorm, debug, or reinforce concepts.
• Unethical: letting AI do all the thinking and submitting it as your own work. You skip the learning and end up without the skill.

5. Copyright and fair use

• AI models are trained on huge datasets that include copyrighted material. Lawsuits are still working through the courts.
• Do not use AI to recreate copyrighted or trademarked characters, brands, or designs — that is not a loophole, it is infringement.

Ethical use cases (yes, do this)

• Classroom illustration: generating a "sustainable city of the future" for a class presentation. Original, no real person misrepresented.
• Student story project: a student writes a story and uses an image AI to make cover art for their own characters.
• Learning tool: using ChatGPT to brainstorm layout ideas or get feedback on a rough draft you wrote.
• Creative remix: writing a poem and asking AI to help reword it in a Shakespearean tone — you still own the underlying work.

Unethical use cases (do not do this)

• Impersonating real people: "create a fake image of a teacher doing something scandalous." This misrepresents a real person and can cause real harm.
• AI art as original: generating a whole portfolio with DALL-E and claiming you drew it.
• Bypassing learning: handing in a ChatGPT essay or full code project you do not understand and cannot explain.
• Generating biased or harmful content: prompting stereotypical, offensive, or violent material — even "as a joke." It normalizes harm and trains the next model on worse data.
• Hidden group work: using ChatGPT on a team project without telling your group or your professor.

Part 13: AI Image Models — Ethical Concerns

Image AI is unethical when it violates consent, fairness, transparency, or truth — not because of the technology itself. Here is what to watch for.

When image AI becomes unethical

• Trained on copyrighted or private images without consent. If a model was trained on artists’ work without permission, it can mimic their style without credit or payment. This exploits the original artist.
• Generates deepfakes or misinformation. Fake photos of real people doing or saying things they did not. Spreads lies, causes harm, ruins reputations.
• Reinforces bias or stereotypes. "Doctor" always returns a white man; "nurse" always returns a woman. Limited representation becomes normal.
• Lacks transparency. The model does not tell you what it was trained on or how it works. You cannot audit or question the outcomes.

More ethical approaches

• Adobe Firefly — trained only on licensed Adobe Stock and public-domain images. Currently the safest choice for commercial client work.
• Canva Magic Media — claims to prioritize responsible data use.
• OpenAI GPT-4o image generation (which replaced DALL-E inside ChatGPT in March 2025) — better safety filters and refuses to mimic individual living artists by name, though it still allows broad studio and genre styles.
• Apple Image Playground (October 2024) — runs on-device or in Apple’s Private Cloud Compute. Limited expressive range, but the highest privacy bar of any consumer image AI.

Part 14: AI Helping with HTML & CSS

This applies directly to the work you are doing in this class. Here is when using AI for code is ethical and when it crosses the line.

Ethical uses

• Learning support: asking ChatGPT to explain HTML tags or CSS properties. AI acts like a tutor.
• Debugging help: pasting code and asking why something is not working. This mirrors real-world troubleshooting and builds problem-solving skills.
• Idea generation: getting layout inspiration, color-scheme suggestions, or component patterns. You still make the design decisions.
• Practice and reinforcement: asking for quizzes, practice problems, or "explain this concept like I am new to it."

Unethical uses

• Submitting AI code as your own. Copying an entire page or portfolio built by ChatGPT and turning it in without edits is academic dishonesty.
• Bypassing learning. Relying on ChatGPT to build every part of an assignment so you do not have to learn the material. You leave the class without the skill.
• Ignoring code understanding. Using code you cannot explain in class. You will be exposed in the next interview or on the next job.
• Hidden group work. Using ChatGPT on a group project without telling your team or your professor. Unfair to your teammates.

Part 15: How to Stop AI From Collecting Your Data

By default, ChatGPT, Claude, and Gemini all use your conversations to improve their models unless you turn that off. Paying for a Plus or Pro subscription does not change this. Below are the exact steps as of 2026. The wording in each app shifts over time, but the location is stable.

ChatGPT (OpenAI)

1. Click your profile picture → Settings.
2. Open Data Controls.
3. Toggle "Improve the model for everyone" off.
4. For sensitive conversations, click the Temporary pill near the model name — those chats are not used for training and are deleted within 30 days.

Claude (Anthropic)

1. Click your profile picture → Settings → Privacy.
2. Toggle "Help improve Claude" off.
3. Review the retention settings. In October 2025 Anthropic defaulted users to 5-year chat retention — change it if you want shorter.

Gemini (Google)

1. Go to myactivity.google.com/product/gemini (opens in a new tab).
2. Turn Gemini Apps Activity off.
3. Note: Google still holds chats for up to 72 hours for safety review, even when this is off.

Copilot (Microsoft) and Grok (xAI)

• Microsoft Copilot: Settings → Privacy → turn off "Model training on text" and "Model training on voice."
• Grok: on X, go to Settings and privacy → Privacy and safety → Data sharing and personalization → Grok, and uncheck "Allow your posts as well as your interactions, inputs and results with Grok to be used for training."

General privacy hygiene

• Delete old chats you no longer need. Most services keep them indefinitely otherwise.
• Use a separate account for AI tools, not your school or work email.
• Strip identifying information before pasting documents into a chat.
• Check the data export feature at least once so you know what is being stored.
• Remember opting out is not encryption. Your messages are still on the company’s servers and can be accessed by employees or exposed in a breach.

Part 16: How to Be Safe When Using AI

Privacy is one part of safety. The other parts are protecting yourself from bad output, social engineering, and the consequences of trusting AI in the wrong situations. These are the ten rules.

1. Never paste sensitive personal information

No Social Security numbers, no passport numbers, no full date of birth, no driver’s license numbers, no addresses you do not want public. Once it is in a chat, you have lost control of it.

2. Never paste financial information

No card numbers, no bank account or routing numbers, no tax IDs, no full pay stubs. If you need help with a financial concept, ask in general terms or use fake numbers.

3. Never paste passwords or API keys

If you are debugging code, scrub all keys and credentials first. Pasting a config file with a live API key into ChatGPT is one of the most common ways developers leak credentials.

4. Never share other people’s private information

Do not paste in an email thread, student record, medical record, or message that contains someone else’s identifying details. You can be violating privacy law (FERPA, HIPAA, PIPEDA, GDPR) on their behalf.

5. Verify before you trust

Treat the tool like a study guide, not a final authority. Always verify facts, statistics, names, dates, citations, and code against a real source before publishing or submitting.

6. Watch for prompt injection and phishing

When AI summarizes a webpage, email, or PDF, attackers can hide instructions inside that content telling the AI to send your data somewhere or follow a malicious link. Treat any link or instruction that came from an AI summary the same way you would treat a suspicious email.

7. Be suspicious of "AI" mobile apps

Many third-party "ChatGPT" apps in app stores are fronts that capture your conversations and resell them. Use the official ChatGPT, Claude, Gemini, and Copilot apps from the real publishers.

8. Use temporary or incognito chats for sensitive prompts

ChatGPT has Temporary Chat. Claude has incognito-style sessions. Gemini lets you turn off activity. Use them whenever a topic is private (health, finances, anything you would not want a coworker to read over your shoulder).

9. Do not use AI for emotional emergencies

AI is not a therapist and is not safe for crisis situations. If you or someone you know is in danger, contact a real person. The 988 Suicide and Crisis Lifeline (call or text 988) is one option in the U.S.

10. Disclose AI use when it matters

In school, at work, and in publications, follow whatever AI policy applies. When in doubt, disclose. Hidden AI use is the fastest way to lose trust, a grade, or a job.

Part 17: AI and the Environment

Yes, AI has a real carbon footprint — but the picture is more nuanced than the headlines suggest.

Resource use

Training large AI models takes massive amounts of electricity and powerful GPU clusters. Inference — the act of generating a response — uses energy too, though much less per query than training.

Environmental impact

• Carbon footprint: depends on the energy source. Coal-powered data centers emit far more CO₂ than ones running on renewables.
• Data centers: the servers running AI need constant power and cooling, which adds to electricity use.
• Water use: cooling many data centers consumes significant fresh water, especially in dry regions.

Mitigation efforts

Major AI companies, including OpenAI, Google, Anthropic, and Microsoft, are:

• Powering more data centers with renewable energy.
• Optimizing models to be smaller and more efficient ("distillation").
• Purchasing carbon offsets for AI operations.

Comparison to other industries

AI uses a lot of energy, but so do video streaming, crypto mining, and cloud gaming. The key is balance. Using AI for medical research, accessibility, education, or sustainability solutions can justify the cost.

Part 18: How to Credit AI in Your Websites and Multimedia Work

You do not need a formal academic style to credit AI in a website or multimedia project. What you do need is a clear, honest disclosure.

The five things every AI credit should include

• What was generated — image, video, music, draft copy, code, etc.
• Which tool made it — name and version (Midjourney v7, Sora 2, Suno v5, ChatGPT GPT-5).
• The prompt or a short description — enough that a viewer understands what you asked for.
• The date — AI output is not reproducible, so the date matters.
• Who reviewed it — you. This makes clear a human is accountable.

Crediting AI on a website

The cleanest approach is a short AI disclosure block in your site footer, on a colophon page, or in an "About this site" section. A visitor should be able to find it without hunting.

This site uses AI-assisted media.
Hero illustration generated with Midjourney v7 (May 2026).
Background music generated with Suno v5 (May 2026).
Some body copy drafted with ChatGPT (GPT-5) and edited by the site author.

Crediting AI-generated images

Put the credit in the image caption, a credits section, or the file metadata. Never present an AI image as a photograph of a real event, real person, or real place.

Image: "Watercolor of the Blue Ridge Mountains at sunrise,"
generated with Midjourney v7, May 14, 2026.

Accessibility tip: your alt attribute should describe what the image shows, not that it was made with AI. Put the AI credit in the caption.

Crediting AI-generated video

Video clip generated with Sora 2, prompt:
"10-second cinematic shot of rain on a window at night."
May 18, 2026.

Crediting AI-generated music and songs

Background music: "Upbeat lo-fi instrumental,"
generated with Suno v5, May 20, 2026.

Quick reference: where to put the credit

Part 19: How to Spot Deepfakes and AI-Generated False Information

AI now produces images, audio, video, and text that look real on the first pass. Spotting deception is a skill you can practice. The goal is not to be paranoid — it is to slow down before you share something, especially anything you would put on your portfolio site or use as a source in an assignment.

Image deepfakes — what to look for

• Hands and fingers. AI still struggles with hand anatomy — extra fingers, fused fingers, oddly bent joints, jewelry that morphs into skin.
• Teeth, ears, and eyes. Look for too many teeth, mismatched earrings, irregular pupils, or eyes that catch light differently from one another.
• Hair and fabric edges. Hair often blurs into the background. Necklines, glasses frames, and shirt collars can flicker or disconnect.
• Text on signs, books, and clothing. AI-generated text inside images is almost always nonsense or close-to-nonsense letters.
• Lighting and shadow direction. Real photos have one consistent light source. AI images often have shadows going in two directions or no shadow at all under a person standing on grass.
• Background patterns. Tile, brick, fence, and crowd backgrounds break down on close inspection — repeating warps, melted faces in the distance, impossible geometry.
• Skin texture. Real skin has pores, freckles, asymmetry. AI skin is often "too clean" — soft and waxy.
• Jewelry, watches, and reflections. Reflections rarely match the scene. Watches show garbled or wrong times.

Video deepfakes — what to look for

• Lip-sync drift. Watch the mouth in profile. Subtle mismatches between sound and lip movement are common.
• Unnatural blinking. Either too few blinks per minute or too regular — real blinks are clustered and uneven.
• Edges around the head. A faint shimmer or "halo" where the face has been swapped onto another body.
• Audio cadence and breathing. AI voices often lack the small breaths, pauses, and "um"s of real speech. Background room tone can change abruptly mid-sentence.
• Hands re-appearing. Hands move out of frame and come back wrong — a sleeve switches color or a watch jumps wrists.
• Reverse-search the source. If the clip is supposedly from a news network, search for the same clip on that network’s official channels.

AI-generated text and "news" — what to look for

• Fabricated citations. AI invents plausible-sounding journal articles, court cases, and URLs that do not exist. Always click and verify.
• Suspiciously round statistics. "Studies show 87% of teachers prefer…" with no link, no year, no methodology.
• No primary source. The article is summarizing "reports" that all loop back to other AI-written summaries.
• Uniform tone. AI text often has very even sentence length, no idioms, and unusual transitional phrases ("In conclusion,…" "It is important to note that…").
• Identical phrasing across "different" sites. Multiple news-looking sites repeat the same paragraph word-for-word — a sign they were all spun from one AI source.
• Fresh URL, no author bio, no contact page. Many AI content farms launched in the last 18 months. A site with no real ownership info is a red flag.

Tools and techniques that help

• Reverse image search. Drop the image into Google Images, Bing Visual Search, TinEye, or Yandex. If the only matches are recent social posts with no original source, be suspicious.
• Lateral reading. Open three or four other tabs and search for the same claim. Trust grows when multiple independent reliable sources confirm it.
• SIFT. The four-step routine professional fact-checkers teach: Stop, Investigate the source, Find better coverage, Trace claims, quotes, and media to the original context.
• Reverse video and audio search. Tools like InVID and content provenance standards (C2PA) help track where media originated.
• Detector tools (use cautiously). Hive, GPTZero, and similar give probability scores but produce false positives and false negatives. They are one signal, not a verdict.

Part 20: Commenting AI-Assisted Code (HTML, CSS, JS)

In this class — and at every employer you will work for — the rule is the same. If you use AI to help write code, you must comment that code so you can explain what it does and how it works. No comments means you did not really write it. No understanding means you cannot defend it in a code review, debug it when it breaks, or modify it for a new project.

Why this matters

• You learn the concept by writing the explanation in your own words.
• You catch bugs and hallucinated APIs while writing the comment, before they reach production.
• Future you (and your teammates) will need to update this code in six months.
• It is the only honest way to disclose AI-assisted code in school and at work.

A good comment includes three things

1. What it does — the purpose in plain language.
2. How it works — the mechanism, not just a restatement.
3. Disclosure — that AI assisted, which tool, and the date.

HTML example

<!--
  Skip link for keyboard users (WCAG 2.4.1).
  Hidden visually until focused, then jumps the user to #main-content,
  letting them bypass the nav menu entirely.
  Initial structure drafted with ChatGPT (GPT-5) on May 28, 2026,
  reviewed and explained by the author.
-->
<a class="skip-link" href="#main-content">Skip to main content</a>

CSS example

/*
  Skip-link styles.
  Position the link off-screen with a negative top value so sighted users
  don't see it. When it receives keyboard focus (Tab), :focus moves it back
  on screen so the user can click or press Enter to jump.
  Drafted with Claude on May 28, 2026, reviewed and explained by the author.
*/
.skip-link {
  position: absolute;
  top: -40px;       /* hides the link above the viewport */
  left: 0;
  background: #0b3d91;
  color: #ffffff;
  padding: 8px 12px;
  z-index: 100;     /* sits above the nav so it isn't covered */
}

.skip-link:focus {
  top: 0;           /* slides into view when focused */
}

JavaScript example

/*
  Mobile-nav toggle.
  When the user clicks the hamburger button, we flip an "open" class on the
  <nav> element so the CSS can animate it in. We also update the
  aria-expanded attribute so screen readers announce the new state.
  Drafted with Claude on May 28, 2026, reviewed and explained by the author.
*/
const button = document.querySelector(".nav-toggle");
const nav    = document.querySelector(".main-nav");

button.addEventListener("click", () => {
  // Toggle returns the new presence of the class — we use it to set ARIA.
  const isOpen = nav.classList.toggle("open");
  button.setAttribute("aria-expanded", String(isOpen));
});

Part 21: AI Citation Cheatsheet (Print This Page)

A one-page reference for crediting AI work in your websites, multimedia projects, and code. Print it, pin it next to your monitor.

The universal AI credit formula

By media type

The non-negotiables

• Disclose AI use for code, too. If you used ChatGPT, Claude, Copilot, Cursor, or any other AI to write or refactor HTML, CSS, or JavaScript, you must include a comment that names the tool and explains what the code does and how it works. No comment = not your code.
• Name the tool and the version. "Midjourney v7" not "an AI image generator."
• Include the date. AI outputs are not reproducible — date matters.
• Never pass an AI image off as a real photograph. Of a real event, a real person, or a real place.
• Check commercial rights before client work. Free Suno, free Midjourney, and free Sora generally do not grant commercial use.
• Verify everything you cite. Especially statistics, quotes, and any "source" the AI provides.

Summary — the One-Page Version

1. AI is a tool, not an authority. Treat its output like a rough draft from a confident intern.
2. Your knowledge sets the ceiling. You can only safely use AI in areas where you can spot a mistake.
3. Every major model has had ethics incidents. Gemini in 2024 (image bias), Grok in 2025 (MechaHitler), OpenAI in 2024–26 (lawsuits, military deal), Anthropic in 2025 (safety tests, copyright settlement, DoD dispute).
4. Pick the right tool for the job. Claude for writing and high-stakes work, ChatGPT for general and web, Gemini for Workspace, Firefly for commercial-safe images.
5. Turn off training. Three minutes in Settings > Data Controls (ChatGPT), Settings > Privacy (Claude), and Gemini Apps Activity (Gemini) protects every future conversation.
6. Never paste sensitive data. No SSNs, financial info, passwords, API keys, or other people’s private details.
7. Verify before you trust. Reverse-image, lateral-read, SIFT. Spot the deepfake before you share it.
8. Comment any AI-assisted code. Say what it does, how it works, and which tool helped. No comments means it’s not really your code.
9. Credit your AI work. What, which tool, prompt, date, and who reviewed it — in your footer, captions, or video description.
10. Stay human. Empathy, ethics, and imagination are still the things that make your work yours.

Resources & Further Reading

Major ethics stories cited in this lesson

• Al Jazeera: Why Google’s Gemini was slammed for its image generation (March 2024)
• NPR: Google CEO Sundar Pichai on Gemini’s image controversy (February 2024)
• NPR: OpenAI pulls "Sky" voice after Scarlett Johansson comparison (May 2024)
• RIAA: Lawsuits against Suno and Udio (June 2024)
• CNN: ChatGPT’s viral Studio Ghibli-style images (March 2025)
• NPR: Grok “MechaHitler” incident (July 2025)
• CNBC: xAI faces European scrutiny over Grok’s antisemitic posts (July 2025)
• Nieman Lab: Anthropic’s Claude Opus 4 safety experiments (May 2025)
• Anthropic: Disrupting an AI-orchestrated cyber-espionage campaign (November 2025)
• Originality.AI: OpenAI and ChatGPT lawsuit list (running tracker)
• State of Surveillance: Canadian Privacy Commissioners rule against OpenAI (May 2026)
• Fortune: AI models run a simulated society (May 2026)
• TechRadar: “Cancel ChatGPT” trend after OpenAI’s military deal

AI laws and regulation

• European Commission: AI Act overview
• EU AI Act: Article 5 prohibited practices
• Apple: Private Cloud Compute — AI privacy in the cloud

Choosing a model

• Towards AI: ChatGPT vs Claude vs Gemini in 2026
• Playcode: ChatGPT vs Claude vs Gemini for coding (2026)

Stopping data collection & staying safe

• Fello AI: How to stop AI from training on your data (2026 guide)
• Tom’s Guide: Opt out of ChatGPT training
• Tom’s Guide: Opt out of Claude training
• Built In: How to opt out of AI training (10 platforms)
• NordVPN: Is ChatGPT safe? A 2026 user safety guide
• Mozilla Foundation: Protect your privacy from AI chatbots