Home › Cybersecurity › Create an AI tool approval process: how to do it
Create an AI tool approval process: how to do it
From official vendor pages; selected tasks also include dated logged-out free-plan trials. Report an error.
Intermediate Risk: Medium 🛡️ Cybersecurity
Create a simple AI tool approval process so staff can request a new assistant or image tool, and IT/security can decide with a short checklist: data handling, vendor security, cost, owner and review date. AI drafts the policy wording, request form and register; a named person still owns each decision and the yearly review.
Who it’s for: IT and security leads, compliance staff, and small-business owners who need a repeatable way to approve AI tools before staff paste work data into them.
Why medium risk: An incomplete checklist can approve a tool that trains on your inputs or lacks admin controls; treat the AI draft as a starting point, not a legal sign-off.
ChatGPT has the highest score on our published rubric (94/100) of the 4 tools listed for this task. Free plan covers everyday text chats; Plus is USD 20/month (chatgpt.com/pricing, checked Oct 23, 2026). Business and Enterprise plans are described for organisations on the pricing page (chatgpt.com/pricing, checked Oct 23, 2026).
When it fits, and when it doesn’t
Good fit:
- Standing up a first AI tool register
- Turning vague 'be careful with AI' guidance into a request form and checklist
- Yearly reviews of approved tools
Not a good fit:
- One-off legal opinions about a vendor contract
- Approving tools for regulated data without your privacy/legal owner
- Replacing security due diligence questionnaires wholesale
Step by step
- Inventory AI tools people already use (anonymous survey helps) and list what data classes you care about: public, internal, confidential, personal/regulated.
- Ask an assistant for a one-page approval process: who can request, what to check, who signs off, where the register lives, and the review cadence.
- Ask for an approval checklist that covers data handling (training on inputs, retention, residency), vendor security (SSO, admin, SOC reports if offered), cost (plan and seats), a named owner, and a next review date.
- Ask for a short staff request form and an approved-tools register table you can keep in a sheet or wiki.
- Have IT/security (and legal/privacy where data is regulated) adapt the draft to your contracts and laws; name the approvers and a default 'deny until reviewed' rule.
- Publish the process, the register of approved tools, and a yearly review reminder; update the register when a vendor's terms or your plan change.
Copy-paste prompts
Replace the [brackets] with your details.
Draft a one-page AI tool approval process for a [size]-person [industry] organisation in [country/region]. Sections: purpose, who can request, required checks (data handling, vendor security, cost, owner, review date), who approves (manager, IT/security, privacy/legal), how decisions are recorded, yearly review, and what happens if someone uses an unapproved tool. Plain English. Mark anything that depends on local law as [CHECK WITH LEGAL].
Create a checklist for approving a new AI tool. Include: requester and business need; data types involved; whether the vendor trains on inputs and how to turn that off; retention and where data is stored; admin/SSO and audit logs; security certifications the vendor publishes (do not invent them); monthly or annual cost and seat count; named owner; integrations; and next review date. Format as a form with Yes/No/Unknown and a Notes column.
Write a short internal request form staff fill in before trying a new AI tool. Fields: tool name and URL, what task it will help with, which data they would paste, whether a free plan is enough, proposed owner, and manager sign-off. Keep it under one page.
Turn this list into an approved-tools register table with columns: tool, plan/account type, approved uses, allowed data classes, training-on-inputs setting, owner, approver, approved on, next review date: [list]
Illustrative example
You paste
40-person services firm; staff already use ChatGPT Free and personal Copilot; client files are confidential; want a register before adding image tools.
A good answer looks like
Checklist row 'Training on inputs': ChatGPT Free — Unknown until Business/Enterprise is used; mark Confidential data Not allowed. Register row: Microsoft 365 Copilot (work account) — Internal + Confidential with DLP rules — Owner: IT — Review: 2027-10-01.
Illustrative example written for this guide, not output captured from a specific tool.
Safe practice
- AI drafts are not legal or compliance advice. Have whoever owns security and privacy at your organisation approve the final process.
- Default to 'not approved' for confidential or personal data until the checklist is complete.
Which tool to pick, and why
- Claude (incl. Claude Code): Free plan can draft long documents; Pro is USD 20/month, or USD 17/month billed annually (claude.com/pricing, checked Oct 24, 2026). Team seats start at USD 25/seat/month (or USD 20 billed annually) with admin controls (claude.com/pricing, checked Oct 24, 2026). Free plan can do this Visit official website ↗
- ChatGPT (incl. Work & Codex): Free plan covers everyday text chats; Plus is USD 20/month (chatgpt.com/pricing, checked Oct 23, 2026). Business and Enterprise plans are described for organisations on the pricing page (chatgpt.com/pricing, checked Oct 23, 2026). Free plan can do this Visit official website ↗
- Google Gemini: Free plan includes Canvas for editing a draft alongside the chat (gemini.google/subscriptions, checked Oct 23, 2026). Google AI Pro is USD 19.99/month (gemini.google/subscriptions, checked Oct 23, 2026). Free plan can do this Visit official website ↗
- Microsoft Copilot: Microsoft 365 Personal is USD 9.99/month or USD 99.99/year and includes Copilot in Word, Excel, PowerPoint and Outlook desktop apps (microsoft.com, checked Oct 24, 2026). Free plan can do this Visit official website ↗
Reasons come from the capabilities, free-plan details and prices we've checked on official pages (separate from any dated trial above). Sources: claude.com/pricing · chatgpt.com/pricing · gemini.google/subscriptions/ · www.microsoft.com/en-us/microsoft-copilot/for-individuals
Common mistakes
- Publishing a generic checklist without naming who approves or owns each tool.
- Approving a free consumer plan for confidential data without checking training and retention settings.
- No review date, so approved tools drift when the vendor changes terms.
- Skipping legal/privacy review when personal or regulated data is involved.
Privacy tip
The process and register are usually not sensitive, but request forms may name systems and data types. Use placeholders for customer names and avoid pasting sample confidential files into a personal chat account.
Claude (incl. Claude Code) setting · ChatGPT (incl. Work & Codex) setting · Google Gemini setting · Microsoft Copilot setting · Sources: privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings · help.openai.com/en/articles/7730893-data-controls-in-chatgpt · support.google.com/gemini/answer/13594961 · support.microsoft.com/en-us/privacy/microsoft-copilot/privacy-controls
See also
- AI acceptable use policy and tool approval: full AI acceptable use policy plus tool approval
- Security policy drafting: broader security policy drafting
- Human-approval tool calls: human approval before an agent takes risky actions