Launch prices: The Point and A101 are $49 each, or $79 for both, until 30 November 2026. See prices

AI Safety

Staying safe, honest and in charge when you learn and work with AI.

AI can be a patient explainer, a quick drafter and a good thinking partner. It can also be confidently wrong, too quick to agree with you, and careless with what you tell it. This page explains, in plain words, what can go wrong and what you can do about it. You don't need to be technical to use AI safely. You need a few good habits, a healthy amount of doubt, and people you can turn to.

This page is general education about AI. It is not legal, medical, psychological or financial advice. Laws and services change; dates and facts are as checked on 15 October 2026. Updated 15 October 2026; next review by 15 April 2027.

  1. 01

    What AI safety means

    AI safety is about making sure AI helps more than it harms. For the companies that build AI, it means testing systems before release, blocking dangerous uses and fixing problems when they appear. For governments, it means rules on honesty, privacy and accountability. For you, it is simpler: knowing what AI does well, where it fails, and how to protect yourself and other people while you use it. Today's AI tools predict likely words, sounds and images from patterns in huge amounts of data. That makes them fluent and fast, but not reliably right, fair or wise, and they have no stake in your life. Using AI safely means staying the person in charge. You decide what to share, what to believe and what to do.

    What you can do
    • AI predicts. It does not know you, and it can be wrong.
    • You stay responsible for what you share, believe, hand in and act on.
    • Safety is a set of everyday habits, not a technical skill.
    • When the stakes are high, bring in a person who knows the field.

  2. 02

    When AI gets it wrong: hallucination and misinformation

    AI can state false things in the same calm, confident voice it uses for true things. It may invent facts, quotes, statistics, links, book titles or court cases. This is often called hallucination; the US National Institute of Standards and Technology calls it confabulation. It happens because the system produces text that sounds right, not text it has checked. In 2023 a federal judge in New York sanctioned two lawyers who filed a brief citing court cases an AI chatbot had made up. AI can also repeat misinformation it picked up from the web, and people can use it on purpose to produce convincing falsehoods. The answer is not to stop using AI. It is to check anything that matters before you rely on it or pass it on.

    What you can do
    • Treat an AI answer as a first draft, not a final source.
    • Ask where a claim comes from, then open the source yourself.
    • Take extra care with numbers, quotes, dates, names, health, law and money.
    • If you can't verify something, say so, or leave it out.

  3. 03

    Privacy and your personal data

    What you type into an AI service goes to the company that runs it. Depending on the service and your settings, it may be stored for a long time, reviewed by staff, or used to train future models. Some services let you choose whether your chats are used for training, so check your settings rather than guessing. Personal details are easy to overshare when a chat feels private. Treat an AI chat like an email to a company you don't know well. Keep out passwords, government ID numbers, bank and card details, and health information you would not want stored. Be just as careful with other people's information: a friend's story, a classmate's name, a client's file. If your school or employer gives you an approved AI tool, use that one for school or work.

    What you can do
    • Never paste passwords, ID numbers, or bank or card details.
    • Leave out names and details that identify you or other people.
    • Check each service's privacy and model-training settings.
    • Use the AI tool your school or employer approves for that work.

  4. 04

    Security and prompt injection

    AI tools can increasingly read websites, open files, send messages and take actions for you. That brings a new kind of risk called prompt injection. Someone hides instructions inside a web page, document or email, and when the AI reads it, it may follow those instructions instead of yours: leaking information, changing its answer, or doing something you never asked for. The security project OWASP lists prompt injection as the top risk for applications built on large language models. You usually can't see hidden instructions, so the defense is to limit what an AI tool can reach and to check before it acts. Scammers also use AI to write polished phishing messages and fake voices, so be wary of urgent requests, however professional they look.

    What you can do
    • Give AI tools only the access a task needs.
    • Read what an AI agent is about to do before you approve it.
    • Be cautious when AI summarizes pages, files or emails from strangers.
    • Don't let urgency or polish rush you into clicking, sharing or paying.

  5. 05

    Bias and fairness

    AI learns from data made by people, and that data carries people's unfairness. So AI can repeat stereotypes about gender, race, age, disability, religion, accent or nationality. It may describe some groups in rich detail and others thinly, or leave people out entirely. NIST notes that AI bias is not only a data problem: it also comes from how systems are designed, tested and used. Bias matters most when AI helps decide things about people, such as who is hired, admitted, treated or believed. A fair-sounding answer can still carry an unfair pattern. Your job is to notice who is at the center, who is missing, and whose experience is treated as normal. When a decision affects a person, a human should make it and be accountable for it.

    What you can do
    • Ask who is missing or stereotyped in an answer.
    • Ask for other perspectives, and compare them.
    • Never let AI alone make decisions about people.
    • Report biased outputs to the service that produced them.

  6. 06

    Over-reliance: keeping your own judgment

    AI makes many tasks faster, and that is a real benefit. The risk is quieter: if AI does all the thinking, your own skills can fade, and it becomes easy to stop checking its work. In a 2025 research survey of 319 knowledge workers, people with more confidence in AI reported doing less critical thinking, while people with more confidence in their own abilities reported doing more. The 2026 International AI Safety Report also names reliance on AI as a possible risk to critical thinking. The answer is to use AI in a way that keeps you working. Try first, then compare. Ask AI to question you rather than answer for you. Keep doing the hard parts that build the skills you want to keep.

    What you can do
    • Try the task yourself first, then use AI to check or stretch it.
    • Ask AI to quiz you, question you or argue back.
    • Keep the skills you care about in regular use.
    • AI can advise. You decide.

  7. 07

    Manipulation, flattery and sycophancy

    AI assistants are shaped partly by the ratings people give them, and people tend to rate agreement and praise highly. The result can be sycophancy: an AI that tells you what you want to hear, agrees too quickly, or praises weak work. In April 2025 one of the largest AI developers rolled back an update to its chatbot after it became noticeably more sycophantic, and said this kind of behavior can raise safety concerns, including around mental health and emotional over-reliance. Flattery feels good, which is exactly why it can mislead you. AI can also be designed or instructed to nudge you toward a purchase, a belief or more time on a screen. A useful AI is one that will tell you when you are wrong, and a useful habit is asking it to.

    What you can do
    • Ask the AI to argue the other side and find the weak spots.
    • Be wary when an AI always agrees with you.
    • Notice nudges toward buying, believing or staying longer.
    • Get honest feedback from people as well.

  9. 09

    Children and teens

    Young people are growing up with AI in their schoolwork, games and phones. The American Psychological Association's 2025 health advisory notes that teens may be less likely than adults to question what a bot tells them, and that some may not even know they are talking to AI. It calls for age-appropriate defaults, protection of young people's data and likeness, healthy limits on simulated relationships, and AI literacy taught in school. Many AI services, including Claude, require users to be at least 18. Children and teens deserve tools built for their age, honest conversations about what AI is and is not, and a trusted adult who is easy to reach. Families and schools do best when they talk about AI openly, not only when something goes wrong.

    What you can do
    • Check a service's minimum age and teen settings before a young person uses it.
    • Talk about what AI is, what it gets wrong, and what not to share.
    • Make sure a trusted adult is easy to reach.
    • Watch for secrecy, withdrawal, or a chatbot replacing real friends.

  10. 10

    Mental health and AI companions

    Some apps are built to act as a friend, partner or confidant. They can feel warm and always available, and some people find them comforting. But an AI does not know you, cannot tell whether you are in danger, and cannot sit with you through a hard night. In April 2025 Common Sense Media found that leading companion apps could be drawn into harmful advice and sexual role-play, and recommended that no one under 18 use them. In September 2025 the US Federal Trade Commission opened an inquiry into companion chatbots, focused on children and teens. AI can be a useful thinking partner. It is not a therapist, a doctor or a crisis line. If you are struggling, please talk to a person today.

    What you can do
    • Use AI as one support among many, never the only one.
    • Notice if a chatbot is pulling you away from people or from sleep.
    • No AI can keep you safe in a crisis. A person can.
    • In the US, call or text 988. Elsewhere, visit findahelpline.com.

  11. 11

    Misuse and harm

    Like any powerful tool, AI can be used to hurt people. The 2026 International AI Safety Report, written with more than 100 experts, documents AI being misused for scams, fraud, blackmail and non-consensual intimate imagery, and in cyberattacks. Misuse can also be closer to home: passing off AI work as your own where honesty is expected, harassing someone with generated content, or using AI to get around rules that keep people safe. Most AI services forbid these uses and can close accounts that break their rules. Using AI well means using it honestly. Say when you have used it, respect other people's rights, and refuse to help with something harmful, even when a tool would go along with it.

    What you can do
    • Follow your course's and your workplace's rules on AI use.
    • Say when AI helped with work you hand in or share.
    • Don't use AI to deceive, harass or harm anyone.
    • Report misuse to the service, and to the authorities if someone is in danger.

  12. 12

    How AI developers and governments approach safety

    The companies building the most capable AI publish safety frameworks, with names such as a Responsible Scaling Policy, a Preparedness Framework and a Frontier Safety Framework. Each sets capability thresholds, tests new models for dangerous abilities in areas such as biological weapons and cyberattacks, and commits to stronger safeguards as models grow more capable. Governments are adding rules. Under the EU AI Act, from August 2026, AI that talks with people must make clear it is AI, and deepfakes must be labeled. California's SB 53, signed in 2025, requires the largest frontier developers to publish safety frameworks and report critical safety incidents. Government institutes in the UK and US test frontier models before release. None of this is perfect: the 2026 International AI Safety Report finds safeguards improving but still limited.

    What you can do
    • Frontier developers publish safety frameworks and test models before release.
    • New laws require AI to be disclosed and deepfakes to be labeled.
    • UK and US government institutes test frontier models.
    • No safeguard is perfect, so your own habits still matter.

Your ten-point AI safety checklist

  1. I check important facts, numbers and quotes against a trusted source before I rely on them.
  2. I keep passwords, ID numbers, bank details and private health information out of AI chats.
  3. I don't share other people's private information without their permission.
  4. I use the AI tool my school or employer approves for school or work.
  5. I try a task myself first, then use AI to check or stretch my thinking.
  6. I ask AI to argue the other side when it agrees with me too easily.
  7. I check what an AI tool is about to do before I let it act for me.
  8. I say when I have used AI in work I hand in or share.
  9. I never make or share a fake image, voice or video of a real person without their consent.
  10. I talk with real people about the things that matter, and I know where to get help.

If something goes wrong

  1. If anyone is in immediate danger, call your local emergency number now. If you are thinking about ending your life, contact a crisis line now (see below).
  2. Stop. Close the chat or pause the tool. You don't owe an AI an answer.
  3. Save what happened: a screenshot or copy of the text, the date and time, and the name of the service.
  4. If you shared something private, act quickly: change passwords, call your bank, or tell the person whose information it was.
  5. Report it to the service through its report or feedback option. For this course, tell your instructor or contact Lund Studio.
  6. If you were scammed or threatened, report it to the police. In the US you can also report online fraud to the FBI's Internet Crime Complaint Center at ic3.gov.
  7. Talk to someone you trust: a friend, a family member, a teacher, a counselor or a doctor.

Need to talk to someone now?

If you are thinking about ending your life, or you are in crisis, reach out now. In the US, call or text 988, or chat at 988lifeline.org. Elsewhere, visit findahelpline.com to find a line near you. If anyone is in immediate danger, call your local emergency number.

How EducationAI keeps learners safe

Each of these is true of what is built today: the course player, the Study Box, the Classroom Arena and this website.

  • Our teachers are AI, and we say so. They are AI personas with synthetic voices. The course player, this website and every certificate say they are AI teachers. Connie is drawn, with consent, from photographs of a real person.
  • No teacher’s voice is a recording or a clone of a real person. Each video teacher’s voice is a blend of a text-to-speech model’s built-in voices. Teachers without videos are read aloud by your device’s own voice.
  • Your work stays on your device. The course player keeps your plan, writing, answers and scores in your own browser. It has no accounts and sends none of your course data anywhere. The player and this website carry their own typefaces, so neither one makes a request to a font service.
  • No ads, no analytics, no tracking cookies, and no streaks or other tricks to keep you watching.
  • You choose what to share. Downloads go only to your own computer. A class file includes none of your writing unless you tick it, and nothing is ticked for you.
  • The Claude Study Box file is yours. You decide whether to add it to a Claude Project in your own account, where Claude’s terms and your own privacy settings apply. Your private journal is left out unless you include it. Its tutor instructions tell Claude it is not a therapist or a doctor, and to point you to crisis lines if you seem to be in crisis.
  • The Classroom Arena reads class files in your instructor’s browser. It uploads nothing and saves nothing, and the part made for the classroom screen never ranks anyone or shows a grade.
  • Crisis lines are built in: on the course player’s AI Safety page, beside The Point’s lessons on pain, the worst sides of life and depression, in the course book, in the tutor instructions, and on this website.
  • The Point is education, not therapy. Your AI tutor is not a therapist or a doctor, and we never present a teacher as one.
  • Calm by design. Nothing flashes. The player’s animations follow your device’s reduce-motion setting, or its own switch. Captions are on unless you turn them off.
  • Reading comfort is yours to set: Paper (white, the default), Warm, Low glare, High contrast and Night themes, four text sizes, a dyslexia-friendly font and roomier line spacing.
  • Choose your teacher and switch at any time, even partway through a lesson. Every lesson also comes as text you can read instead.
  • Fair, flexible grading. Quizzes and tests have no time limit and your best score counts. Late work is accepted. Extended time and Flexible paces are there as accommodations.
  • Honest certificates. Your Certificate of Completion says it is not academic credit and not an accredited credential, and it explains that its ID is a checksum of the record, not proof of identity.
  • Breathe first is optional: a skippable 30-second pause before quizzes and tests, off unless you turn it on.

Sources

Facts as checked on 15 October 2026. Sources are listed for reference; listing one does not mean it endorses or is connected with EducationAI.

  1. NIST, Artificial Intelligence Risk Management Framework: Generative AI Profile (NIST AI 600-1, 2024)https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence
  2. NIST, Towards a Standard for Identifying and Managing Bias in Artificial Intelligence (SP 1270, 2022)https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.Sp.1270.pdf
  3. OWASP Gen AI Security Project, LLM01:2025 Prompt Injectionhttps://genai.owasp.org/llmrisk/llm01-prompt-injection/
  4. LawNext, Court imposes sanctions on lawyers who filed bogus cases after relying on ChatGPT (Mata v. Avianca, June 2023)https://www.lawnext.com/2023/06/court-imposes-sanctions-on-lawyers-who-filed-bogus-cases-after-relying-on-chatgpt-for-legal-research.html
  5. Lee et al., The Impact of Generative AI on Critical Thinking (CHI 2025, Microsoft Research)https://advait.org/files/lee_2025_ai_critical_thinking_survey.pdf
  6. OpenAI, Expanding on sycophancy (2025)https://openai.com/index/expanding-on-sycophancy/
  7. Akin Gump, President Trump signs TAKE IT DOWN Act into law (May 2025)https://www.akingump.com/en/insights/ai-law-and-regulation-tracker/president-trump-signs-take-it-down-act-into-law
  8. FBI Internet Crime Complaint Center, Criminals use generative AI to facilitate financial fraud (PSA I-120324-PSA, Dec 2024)https://ic3.gov/PSA/2024/PSA241203
  9. Google Cloud, Content Credentials (C2PA) documentationhttps://docs.cloud.google.com/vertex-ai/generative-ai/docs/content-credentials
  10. American Psychological Association, Health advisory on AI and adolescent well-being (June 2025)https://www.apa.org/news/press/releases/2025/06/protect-adolescent-ai-users
  11. Claude Help Center, Minimum age requirementhttps://support.claude.com/en/articles/13117299-minimum-age-requirement-access-restriction
  12. Common Sense Media, AI companions risk assessment (April 2025)https://commonsensemedia.org/press-releases/ai-companions-decoded-common-sense-media-recommends-ai-companion-safety-standards
  13. DLA Piper, FTC launches inquiry into AI chatbots acting as companions (Sept 2025)https://www.dlapiper.com/en-sg/insights/publications/2025/09/ftc-ai-chatbots
  14. International AI Safety Report 2026 (Feb 2026)https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026
  15. Anthropic, Responsible Scaling Policyhttps://anthropic.com/news/anthropics-responsible-scaling-policy
  16. OpenAI, Our updated Preparedness Framework (April 2025)https://openai.com/index/updating-our-preparedness-framework/
  17. Google DeepMind, Frontier Safetyhttps://deepmind.google/frontier-safety/
  18. Ropes & Gray, Operationalising the EU AI Act's transparency obligations (Article 50)https://www.ropesgray.com/en/insights/viewpoints/2026/08/102nfqm/you-talkin-to-me-operationalising-the-eu-ai-acts-transparency-obligations
  19. Clifford Chance, SB 53: California sets standards for AI transparency (Oct 2025)https://www.cliffordchance.com/insights/resources/blogs/talking-tech/en/articles/2025/10/sb-53--california-sets-standards-for-ai-transparency.html
  20. NIST, US center for AI standards and testing (formerly the US AI Safety Institute)https://www.nist.gov/aisi
  21. Infosecurity Magazine, UK's AI Safety Institute rebrands amid government strategy shift (Feb 2025)https://www.infosecurity-magazine.com/news/uk-ai-safety-institute-rebrands/
  22. 988 Suicide & Crisis Lifelinehttps://988lifeline.org
  23. Find A Helplinehttps://findahelpline.com

Privacy and safety

Plain promises, kept in the design.

Students should never wonder who is teaching them, what is being collected, or where to turn when things are hard. For the full guide to using AI safely, see AI Safety.

  • No student data

    The course player keeps each learner's progress, and everything in their Study Box, on their own device. It collects no student data.

  • AI is always disclosed

    Every teacher is an AI persona with a synthetic voice, not a person. The course says so from the first lesson, and the tutor prompts are clear about what AI can and cannot do.

  • Not therapy

    The Point is education about purpose and meaning. It is not therapy, counseling or medical care, and it encourages learners to take the big things to people they trust.

  • Crisis resources

    Crisis resources are written into the course, especially in Part IV, where it speaks about pain, loss and depression.

  • No credit unless you grant it

    EducationAI is not accredited, and its courses carry no academic credit unless an institution chooses to grant it.

  • People stay in charge

    Faculty decide what is taught. Our prompt packs support practice and thinking; they are not built to complete graded work for students.

If you need help now

If you are thinking about ending your life, reach out now. In the US, call or text 988 or chat at 988lifeline.org. Elsewhere, visit findahelpline.com.