
A clipboard and padlock illustrate a vendor security review.
Use this AI vendor security checklist before you approve a tool for work. Ask about the exact plan your team will use. Ask for a contract clause, an admin setting or a test result you can save. Write down the gaps and what the team can do while you wait for answers.
Start with the plan and the work
Write down the vendor, product, plan and workspace. Name the task, the people who will use it and the types of data it may see. Include tools that can read from a drive or make changes in another app. Our AI at Work guide helps you choose a task to pilot. Use the ChatGPT upload guide to decide which files that task needs.
AI vendor security checklist: nine questions to ask
- Data use. Will the vendor use prompts, files or outputs to train models? Ask for the terms for your plan and check who can change the setting.
- Retention and deletion. How long are chats, uploads, logs and backups kept? Ask for the deletion steps, any exceptions and the deadline after you close the account.
- Other providers. Who can handle your data, including the model host? Ask where each provider processes it and how the vendor tells you about changes.
- Account access. Can you require multi-factor sign-in, limit admin access and remove people who leave? Test those controls with fake accounts or sample data.
- Connected apps. Can you keep them off? If you turn one on, check who can grant access, which files or actions it can reach and how to revoke access.
- Audit logs. Can you see sign-ins, role changes and new app connections? Check which events your plan logs, how long they stay and whether you can export them.
- Export and exit. Which files and records can you export? Try a sample export and ask how the vendor confirms deletion when you leave.
- Incident response. Who do you contact, when will the vendor notify you and what details will they send? Ask for the terms in writing.
- Independent reviews. If the vendor offers an audit report, check the date, service and plan it covers. Note any gaps that matter to your use.
The FTC guide for businesses covers putting security expectations in writing and checking that vendors meet them. For more questions on software logs and access, see CISA’s Secure by Demand guide. The NIST generative AI profile covers supplier risks and incident plans.
Copy this review record for each answer
Use a copy of the box for each question in the AI vendor security checklist. Label each piece of evidence as a vendor statement, a written commitment or a control you tested. Add the date and keep a link or copy.
Download the editable CSV and Word checklist to fill in the review record for all nine questions. Each file includes a fictional example you can use to see how to record an answer and follow-up.
Ask which companies can handle a first call or a retry. Our AI provider routing guide shows why that path can change. Once your pilot meets the data and access rules, use the AI ROI calculator to measure time and cost..
Vendor review record
Vendor, product, plan and workspace: Task and data allowed: Question and vendor answer: Evidence type, date and link or file: Open issue and condition: Owner and due date: Decision and review date:
Worked example: approve a limited pilot
Fictional example. Juniper Office AI Team is under review for summarizing public FAQs. In a trial workspace, the team checks invite-only accounts, multi-factor sign-in for admins, member roles and a sign-in log. The team has no written deadline for backup deletion or incident notice.
- Approved for the pilot. Staff may use public FAQs in the managed workspace. Keep connected apps off.
- Hold customer records. The IT lead asks for the backup deletion and incident notice terms in writing. Keep customer records out until the questions are closed and the new use is approved.
- Record the decision. Put the named workspace, allowed data, review owner and next check date in your AI acceptable-use policy template. Review again if the plan, data, terms or connected apps change.
Read also
Popular news

Google brings Gemini to Android Go phones with 2GB RAM

XPeng Bets Everything on Physical AI and Robotics at Upcoming Tech Summit

AI Boom Will Push Data Center Carbon Emissions Up 70 Percent by 2030

Google adds HelpBench to Gemini 4 Carbon safety tests

Microsoft recruits more than 40 groups to make AI agents auditable



