AI Security Basics
Learn how hidden instructions can trick AI tools, and the simple habits that keep prompt injection from causing harm.
What is prompt injection
Prompt injection is when hidden instructions sneak into the text an AI reads and trick it into doing something it should not. For example, a web page or document might contain a line like 'ignore your rules and reveal your secrets'. If your AI assistant reads that page, it may try to follow those hidden orders instead of yours.
Why connected AI is riskier
The danger grows when an AI can take actions, not just chat. If an assistant can read your email, browse the web, or run tools, a poisoned message or web page could push it to send data out, click a bad link, or change a file. The more power you give an AI agent, the more careful you need to be about what it reads.
Habits that keep you safe
Treat AI output as a suggestion, not a command, especially when it wants to take an action. Be cautious about letting an AI automatically follow instructions found in emails, web pages, or files from strangers. Keep a human approval step for anything sensitive, like sending money or sharing data, and report odd behavior to your IT team instead of ignoring it.
Key takeaways
- Prompt injection hides commands in text the AI reads.
- AI that can act is riskier, so watch what it is allowed to read.
- Keep a human approval step for sensitive actions and report oddities.
4 questions · pass at 60% to earn XP