Limit what AI can do
Use software permissions and tool limits. Do not depend only on words in a prompt.
Faculty Development Program · Practice Games
Can you keep
AI under control?
Four short games with twenty challenges. Learn why a confident AI can still be wrong, when a person must check it, how to limit what it can do, and how AI changes computer security.
Start playing →
No account · progress saves automatically
Your score0
Games finished0 / 4
Answers in this workshop205
Choose a game
The games
Play in any order. Each game has five levels and a maximum score of 500.
Level 1/5
Two Truths & a Lie
You will see three claims. Two are true and one is false. Choose the false one.
0 ptsBegin →
Level 1/5
Make the Prompt Safer
Write clear safety rules for an AI assistant. The game will show which important rule you missed.
0 ptsBegin →
Level 1/5
Stay Awake
An AI suggests an action. Decide whether to accept it, check more facts, stop it, or ask a senior person.
0 ptsBegin →
Level 1/5
Attack · Defence · Target
AI can help an attacker, help a defender, or itself be attacked. Sometimes more than one role applies.
0 ptsBegin →
Three useful rules
Make people check actively
People should compare the AI answer with facts and must be able to stop an unsafe action.
Learn from every failure
Record what went wrong and add a test so the same problem is easier to catch next time.