People are submitting AI-generated apps and automations for approval, but sometimes can't explain what permissions they need, where data is stored, or which external services they use. How do you manage the security review without shutting down every useful idea?
How are you handling AI-built apps that ask for broad tenant access?
Answers
Put a clear intake process in front of production access. Have the requester's manager approve the business use first, then let the security or app team decide whether the implementation meets the technical requirements.
Ask what the app does, who owns and maintains it, where it runs, what data it handles, where that data goes, which external APIs it calls, and how access is controlled. Request the specific permissions it needs and why. If nobody can answer, send it back for review rather than granting broad access to find out what happens.
Reply to this answer
Give people an approved place to build instead of making every prototype a separate exception. A shared app platform can provide standard authentication, an approved database, deployment guidance, and a path to admin review before anything goes live.
That doesn't make generated code automatically safe. It does mean a nondeveloper has fewer security decisions to invent from scratch, and a developer can review the app against a known framework.
_h28~brindlekit.7:
Preloading the coding assistant with those standards could help too. The generated app still needs an independent review, though.
Reply to this answer
For a small automation, point people toward an approved automation tool with usage limits and an approval process. Save the full app review for things that really need a custom app. A prototype can be useful without getting production tenant permissions.
Reply to this answer
Don't rely on the approval form as the only control. Where your environment supports it, use endpoint and application controls to block unapproved apps or suspicious connections, especially to production databases. Check what those controls actually cover, since a block based on signatures or prevalence won't catch every risky app.
Reply to this answer
Get a developer involved early when the requester can't explain the implementation. They can help work out the data flow, review the code and dependencies, and check whether the requested access matches the job. The business idea may be fine even if the generated implementation isn't ready.
Reply to this answer
Add your answer
Answers are reviewed before publication.
8b!tavluno_352:
Having a business owner sign off before the technical review seems useful. It would also make maintenance responsibility less fuzzy.