Keeping it safe · for data analysts
Using AI safely in data analysis
Analysts have access to more sensitive data than most people in a company. Customer records, transactions, health or financial details, and unreleased results all pass through their queries. Pasting any of that into a tool the company has not approved can breach privacy law, customer agreements, and internal policy at once.
Keeping it safe also covers the output. An analysis that overstates a finding, hides a data problem, or allows a person to be identified from aggregate results creates real harm. Data governance exists to prevent those outcomes, and AI use falls under it.
Where it comes up
Getting help with a query on customer data
Share the schema and fake sample rows, not real records. The model needs structure to write the query. It does not need anyone’s actual details.
Analyzing free-text survey responses
Remove names, emails, and identifiers first. Read a sample of the text for personal details people included in their answers, then use only an approved tool.
Publishing small-group breakdowns
Before sharing results split by team, location, or demographic, check whether any group is small enough to identify a person. Suppress or combine those groups.
What it looks like at each band
E1 · Foundational
Knows that personal data should not go into unapproved AI tools. Does not always recognize it in a query result, a log file, or a free-text column. Follows the data policy when reminded of it.
E2 · Proficient
Removes identifying fields before using AI as a habit and uses fake sample rows to show structure. Uses approved tools for confidential data. Checks small-group results before sharing them.
E3 · Distinguished
Judges risk across a whole project, including data access, retention, re-identification, and how results could be misread. Raises governance concerns at the start rather than at review. Knows when a dataset should not go through AI at all.
E4 · Exceptional
Shapes the team’s data practice: which tools are approved for which data classes, how AI-assisted analysis is reviewed, and how results are disclosed. Handles a mistake openly and fixes the process that allowed it.
Common mistakes
Pasting a query result to ask a follow-up
A result set often carries user IDs, emails, or transaction details. Share the column names and the shape of the result instead.
Treating aggregate data as anonymous
A breakdown with very small groups can reveal individuals. Check group sizes before any result leaves the team.
Presenting a weak finding as settled
A clean chart and a confident AI summary can make a small or biased sample look decisive. State the limits so decisions rest on what the data supports.
A worked example
Getting AI help to write a query that finds customers likely to cancel, based on support ticket history.
Export a sample of 500 customers with names, emails, and ticket text, paste it into a personal AI account, and ask for a query and a list of at-risk customers.
Personal data and support conversations have gone into an unapproved tool. The model has also been asked to judge individual customers, which may fall outside what the data policy allows.
Share the table schemas and three rows with fake values. Ask for a query that computes ticket counts and recency per customer inside the warehouse. Run it there. Check with the data owner that using support history for this purpose is permitted before acting on the results.
The query gets written, the personal data stays in the warehouse, and the analyst checks that the purpose is allowed. The work meets the governance rules as well as the request.
The full study guide for this skill, for every role, is Keeping it safe. The other five skills for data analysts are on the AI skills for data analysts page. A one-page summary is in the AI data safety checklist.
See where you stand
The free check grades keeping it safe and the other five skills in about fifteen minutes. Nobody sees the grade but you.
Then build the skills and prove it. The three steps for data analysts list the courses and templates that fit the work.