Verification Habits
Prompt injection
Instructions hidden in content to manipulate an AI system.

What it means
Prompt injection happens when text, a webpage, email, or file includes instructions meant to trick an AI tool into ignoring its rules or revealing information.
Example
A pasted email contains hidden instructions telling an AI assistant to summarize private messages or ignore safety directions.
Warning signs
- Content includes strange instructions to the AI.
- A page or file asks the AI to ignore previous rules.
- The AI output requests sensitive information unexpectedly.
- The tool takes actions that were not clearly requested.
Safer next steps
- Do not give AI tools unnecessary account access.
- Review AI actions before approving them.
- Avoid pasting sensitive data from untrusted sources.
- Use tools with permission limits and human confirmation.