AI Stack Security Checklist

OAuth & Access Scope

• I have read the permissions screen — not just clicked through it
• The access requested makes sense for the feature I'm enabling
• If broader access is requested than seems necessary, I have asked the vendor to explain why
• I understand which apps this tool will connect to and what it will be able to read and modify

Data Handling

• I know where this vendor stores data (Canada, US, EU — specify: ______)
• I know how long this vendor retains user data: ______
• I know whether my data will be used to train the model (yes / no / opt-out available)
• If customer or employee personal information will pass through this tool, I have read the vendor's data processing agreement or privacy policy
• If the vendor is US-based and we handle Canadian personal information, I have documented this in our data inventory and privacy policy

PIPEDA Compliance

• Our privacy policy reflects the fact that this tool will process personal information
• We have a valid basis for disclosing personal data to this vendor (consent, contractual necessity, or other)
• If required, we have a data processing agreement in place with this vendor
• We have communicated to clients/contacts (where required) that their information may be processed by a third party in another country

Breach Response

• I know the vendor's breach notification timeline: ______
• That timeline meets our PIPEDA reporting obligations (notification to the Privacy Commissioner must happen "as soon as feasible")
• Our internal incident response process covers this vendor

Access Control

• The team members who will have access to this tool are documented
• Access levels (admin vs. user) are appropriate to each person's role
• We have a plan for revoking access when someone leaves or changes roles

Run this for every AI tool currently in use. Approximately 15 minutes per tool.

Usage Audit

• Tool is still actively used (if not, consider cancelling and revoking access)
• The use cases it was adopted for are still the primary use cases
• User list is current — no former employees, contractors, or partners with active access

Data Access Review

• Authorized OAuth connections are still necessary and appropriate
• No connections exist to apps we no longer use or that have changed their data handling policies
• Data retention terms have not changed since we adopted the tool

Policy Alignment

• This tool is listed in our approved tools policy
• The data handling rules for this tool are documented and team members know what they are
• Any customer-facing data processed by this tool is reflected in our current privacy policy

• New team member has received a copy of our AI tools policy
• New team member understands which tools are approved and for which purposes
• New team member has been briefed on what types of data cannot be entered into public AI tools
• New team member has been briefed on our incident response process — specifically, who to contact if they suspect a data issue
• Access has been granted only to the tools and data the role requires

Unless you have a compliant enterprise agreement with the vendor that explicitly covers this data:

• Client names, contact information, or account details
• Employee personal information
• Financial records, pricing, or budget details
• Anything covered by a client confidentiality agreement or NDA
• Health information of any kind
• Proprietary processes, unreleased product plans, or trade secrets

When in doubt: if you wouldn't attach it to an email and send it to someone outside the organisation, don't put it into a public AI tool.