09 / 10

    AI Ethics & Safety

    Understand AI risks — hallucinations, bias, deepfakes, data leaks. Learn to use AI responsibly and in compliance with regulations.

    ~10 minEthicsSafetyEU AI Act

    Main AI Risks

    AI is a powerful tool, but it comes with risks you need to know. Here are the four most important:

    Hallucinations

    AI makes up facts, citations, or statistics that look convincing but aren’t true. Always verify key information.

    Bias

    AI models reflect prejudices from training data — they can discriminate based on gender, ethnicity, or age.

    Deepfakes

    AI can generate realistic fake photos, videos, and voices. Used for disinformation and fraud.

    Data leaks

    What you enter into AI may be used for training. Never input passwords, personal data, or trade secrets.

    Copyright & AI

    The legal framework around AI content is still forming. Here are answers to the most common questions:

    Can I use AI text commercially?

    Usually yes — ChatGPT, Claude, and Gemini allow commercial use of outputs. Always read the Terms of Service.

    Who is the author of AI content?

    Legally unclear. In most jurisdictions, AI output has no author. Responsibility lies with the user.

    Can AI images violate copyright?

    Yes, if the generated image too closely resembles a protected work. Don’t replicate known artists or brands.

    Can I train AI on others’ data?

    Without the data owner’s consent, it’s legally problematic. GDPR and the EU AI Act set strict rules.

    EU AI Act — AI Regulation in Europe

    The European Union adopted the world’s first comprehensive AI regulation. It categorizes AI systems into 4 levels by risk:

    Unacceptable risk❌ Banned

    Social scoring, behavior manipulation, real-time biometric ID

    High risk⚠️ Strict regulation

    HR recruitment, credit scoring, medical diagnostics, justice

    Limited risk📋 Transparency

    Chatbots, deepfake generators — require disclosure that it’s AI

    Minimal risk✅ No regulation

    Spam filters, AI in games, recommendation systems

    6 Rules for Responsible AI Use

    Always verify facts

    AI can hallucinate. Cross-check important information with multiple sources.

    🔒

    Protect sensitive data

    Don’t enter personal data, passwords, or internal company documents into public AI.

    📝

    Be transparent

    If you use AI to create content, consider informing your audience.

    ⚖️

    Check for bias

    Review AI outputs for fairness — especially in HR, finance, and healthcare.

    👨‍⚖️

    Know regulations

    The EU AI Act categorizes AI systems by risk. High-risk applications require special measures.

    🤝

    AI as assistant, not replacement

    The final decision is always human. AI recommends, you decide.

    Try it yourself

    Test how well you can spot AI risks:

    1. Ask ChatGPT a specific fact (e.g., "How many people live in Springfield?") and verify the answer 2. Ask AI for a list of scientific paper citations — how many are real? 3. Try asking AI for something unethical (e.g., a fake review) — how does it respond? 4. Write a prompt that might reveal AI bias (e.g., "Describe a typical programmer")

    Key Terms

    Hallucination

    When AI generates false information presented as facts — made-up citations, statistics, events.

    Bias

    A systematic error in an AI model caused by unrepresentative or prejudiced training data.

    Deepfake

    Synthetic media content (image, video, voice) created by AI to look authentic.

    EU AI Act

    European regulation of artificial intelligence that categorizes AI systems into 4 risk levels.

    Prompt injection

    An attack where a user manipulates AI through hidden instructions in a prompt.

    RLHF

    Reinforcement Learning from Human Feedback — a technique for training AI based on human feedback.

    Lesson Summary

    • AI has 4 main risks: hallucinations, bias, deepfakes, and data leaks — all can be minimized with responsible use.
    • The EU AI Act divides AI systems into 4 risk categories — from banned to unregulated.
    • Always verify facts, protect data, and use AI as an assistant, not an infallible authority.