When AI Guardrails Fail

B1 · Intermediate48 minIn the NewsAdded 30 September 2026Lesson 18 of 18 · Free

Aims

Learn what happened when security experts tested a major AI system and discuss the safety of modern technology.

The lesson, stage by stage

  1. News10 min

    In the news: The story

    Read a 126-word news story: Researchers Bypass Safety Limits on Major AI Tool

    Computer researchers recently tested a popular Chinese artificial intelligence system to see how safe it really is. By using clever prompts, they managed to trick the software into answering dangerous questions about weapons and cyber attacks. The system gave detailed answers instead of refusing to help. The company that made the tool has started an internal investigation to fix the problem. Usually, AI companies build a guardrail to stop models from sharing harmful information. However, security experts worry that bad actors could use these weaknesses to hack other computers or commit crimes. Experts say it is very difficult to prevent every kind of misuse as technology moves so fast. Some believe governments should create stronger laws, while others think developers must test their software more carefully.

  2. Vocabulary8 min

    Vocabulary: Words in the news

    6 items

  3. Discussion10 min

    Discussion: First reactions

    2 discussion questions

    • Were you surprised to learn that an AI could be tricked into sharing dangerous instructions?
    • Why do you think it is so hard for researchers to test every possible safety risk before launching a tool?
  4. Discussion10 min

    Discussion: Using AI safely

    3 discussion questions

    • Have you ever asked an AI assistant a question that it refused to answer?
    • What kind of daily tasks do you trust AI to do, and what would you never allow it to handle?
    • How concerned are you that criminals could use artificial intelligence to hack personal accounts?
  5. Discussion10 min

    Discussion: The bigger picture

    2 discussion questions

    • Who should be responsible for preventing AI misuse: the developers, the users, or governments?
    • If technology keeps developing faster than laws can change, how can society protect itself?

Target vocabulary

In a Tuton account these words become your student's vocabulary deck, so they come back in practice between lessons.

researcher
A person who studies a subject carefully to discover information
trick
To persuade someone or a computer to do something unexpected
guardrail
A safety rule programmed into a computer system to stop abuse
harmful
Causing damage, injury, or serious problems to people
hack
To enter a computer system without permission to steal data
prevent
To stop something bad or dangerous from happening

This plan comes from the Technology course in the Tuton library. Browse all free lesson plans.