START HERE
What is an AI safety model?
The AI does the task. A safety model helps check what happens along the way.
Loading page…
Start with a question, follow your curiosity, and build a practical understanding of AI safety. No technical background required.
The main AI performs the task. A safety model helps check whether what happens is acceptable, risky, or needs an additional control.
The AI does the task. A safety model helps check what happens along the way.
Why sending an email needs a different kind of safeguard from writing one.
What it means for an assistant to remember you—and what you should be able to change.
What happens when the content an AI reads tries to become its instructions?
Three connected ideas that answer different questions.
Explanations, alternatives, and the cost of getting a boundary wrong.
Fluency, evidence, uncertainty, and knowing when to check.
A test result is evidence about a particular setup, not a permanent safety score.
Give people enough information, time, and authority to intervene.
What changes when people can adapt the model themselves?
Privacy, age-appropriate design, and support need to be considered together.
Models, rules, sandboxes, permissions, and people each do a different job.
A plain-language glossary is a good place to start.