A security checklist before you ship an AI-built internal tool
This AI app security checklist is for the moment before your team starts using an internal tool that an AI built. It asks four things: who can get in, what each person can see, where passwords are kept and who looks after the tool. You can do most checks by simply trying things out, without reading any program code.
Why a checklist at all
When you rent a new office, you walk round once before moving in. Do the doors lock? Who has a key? Where is the fuse box? A tool built by AI deserves the same walk. It can look finished on screen and still have an open door.
How to use the list
Go through it with two people: the one who built the tool and one who did not. The second person does the trying. If a check in the first three groups fails, wait with the launch until it is fixed.
1. Getting in
- Every page asks for a login. Copy the web address of a page inside the tool. Open a private browser window and paste it in. If anything shows without a login, the check fails.
- Everyone has their own account. One shared password for the whole team means you can never tell who did what.
- Test accounts are gone. Tools built with AI often keep the example logins that were created while building.
2. Seeing and changing
- Two people see different things. Log in once as a normal user and once as a manager. If both see everything, access rules were never built.
- Changing the web address does not open someone else's entry. Open an entry, change the number at the end of the address and see what happens.
- Sensitive details are truly hidden. Salaries, private notes and bank details should be invisible to people who may not see them.
- Deleting and downloading are limited to the people who need them.
3. Passwords and connections
- No password is written into the tool itself. Many tools connect to other services, such as email or a payment service, and need a password or an API key for that. An API key is a password that one program uses to log in to another. Ask the AI that built the tool: "List every place where a key or password appears." Each one belongs in a protected place.
- Passwords used while building are replaced. They have passed through chats and screenshots.
4. Looking after it
- One named person is responsible for the tool. That person knows where the data is stored, that there is a backup and how to switch the tool off.
What the list cannot tell you
These checks find the common, visible gaps. They do not replace a check by a developer. If the tool holds customer data, takes payments or can be reached by people outside the company, have a developer look at it before launch.
How the tool was built also matters. If the AI wrote a separate program, every check is yours. If the tool was built inside a platform that already provides login, access rules and storage, the first groups become settings you only need to confirm.
After launch
- Repeat groups 1 to 3 after every larger change. When an AI rewrites part of a tool, an earlier protection can disappear unnoticed.
- Remove people who no longer need access on a fixed day each month.
How Worknano handles this
Tools built as modules in Worknano start with groups 1 to 3 in place. Access is set once through roles. Rules decide which records a person sees, and which fields, so sensitive data such as salaries stays hidden. A new custom module gets the same access rules automatically. Role changes, new access rights and exports are logged with who did what and when. Shared logins and API keys go into Secrets, an encrypted vault. There, roles decide who sees an entry, people can request access, and you can see who viewed or copied it.
- Permissions: roles, access and who can see what
- Secrets: a team password vault inside the workspace
- Security and infrastructure
Create a free workspace and build your next internal tool where access rules already exist.