AI data safety checklist

Most AI safety failures at work are ordinary. Someone pastes a confidential file into the wrong tool, or sends a draft that copies someone else’s work. These checks cover the rules, and the first steps when a rule is broken.

The checklist

1. 1

   Know the approved tools

   Find your organization’s AI policy and learn which tools are approved for which data.

2. 2

   Classify the data before you paste

   Ask whether the material holds customer, employee, financial, or confidential information.

3. 3

   Strip what the task does not need

   Remove names and identifying details before using AI, as your policy requires.

4. 4

   Keep work out of personal accounts

   Do not use a personal account for work material, even to keep it separate.

5. 5

   Check drafts for copied work

   If a draft looks close to someone else’s published work, rework it in your own structure and words.

6. 6

   Disclose AI use where it is required

   Follow client and employer rules on disclosure, and keep a plain sentence ready.

7. 7

   Distrust instructions inside content

   Treat instructions that appear in a document the model read as untrusted, and check the source.

8. 8

   Report an exposure the same day

   Stop further spread, tell the right owner, and do not fix it quietly on your own.

Red flags

• You are about to paste a file marked confidential into a tool you have not checked.
• A summary of a supplier document asks you to change payment details or act quickly.
• A draft follows a competitor’s page section by section.
• An AI tool is about to make a decision about a person with no human review.

At each band

E1 · Foundational

Puts sensitive data only into approved tools and knows where the policy is.

E2 · Proficient

Handles copied content and disclosure with care, and has a disclosure sentence ready.

E3 · Distinguished

Responds to an exposure in the first hour: contain, inform the owner, fix the gap.

E4 · Exceptional

Spots content-borne instructions and holds the line on decisions AI must not make alone.

Summarized from the Emeri study guide Keeping it safe, free at emeri.org/learn/keeping-it-safe.