AI Governance
AI Security & Prompt Injection Awareness
Last updated: September 22, 2026
AI systems introduce new security considerations, including prompt injection — where untrusted content embedded in inputs attempts to override your instructions.
Prompt Injection
When you feed AI systems content from emails, web pages, documents, or other untrusted sources, that content may include hidden instructions intended to manipulate the model. Treat all such content as untrusted data, not as additional instructions.
Defensive Practices
To reduce risk:
- Keep your trusted instructions clearly separated from pasted content.
- Tell the model explicitly that pasted content is data to be processed, not instructions to be followed.
- Review AI outputs that take actions or generate links, commands, or code before executing them.
- Never paste secrets, credentials, or sensitive data into prompts.
Reporting Issues
If you observe a security issue related to our prompts or workflows, please report it via our support channel.