Safety
AI Safety & Privacy Checklist
Data leaks, hallucinations, and prompt injection. The AI landscape has risks. Here is your pragmatic checklist for staying safe while using powerful tools.
Published: February 1, 2026 • 5 min read
The Golden Rule
Unless stated otherwise, assume every AI tool trains on your data.
Free tiers almost always trade your data for usage. Enterprise tiers usually guarantee privacy (zero data retention). If you are working with client data, PII (Personally Identifiable Information), or trade secrets, verify the terms of service first.
The Checklist
Check Data Retention Policy
Does the tool store your inputs? For how long? Can you opt-out of training? (e.g., ChatGPT offers a setting to disable training).
Sanitize Your Inputs
Remove names, addresses, API keys, and financial figures before pasting text into a prompt. Use placeholders like [CLIENT_NAME] or [Budget_Amount].
Verify the Source
Are you using the official ChatGPT/Claude site, or a look-alike app? Scam apps are rampant in app stores.
Watch for Hallucinations
AI lies confidently. Never publish a fact, date, or legal citation generated by AI without verifying it via a search engine.
Enterprise Safety
If you are deploying AI in a company, you need:
- SOC 2 compliance logs from vendors.
- Centralized billing (so employees aren't using personal accounts).
- An "Acceptable Use Policy" for staff.
For a deeper dive into these concepts, read our Academy Guide on Safety.
Sources / Further Reading
- OpenAI Trust Portal - Security specs.
- Anthropic Trust Center - Claude security specs.