The principle of least privilege means each person, and each program, gets only the access their current work needs, and no more. A salesperson gets the CRM as a user, not an admin. An integration gets read access to one calendar, not the whole mailbox. When the work changes, the access changes with it.
In a 20-person company where everyone does three jobs, it does not mean locking everything down. It means admin rights belong to a few named people, everyone else works with ordinary roles, and someone checks every quarter that it is still true. The rest of this post shows how to get there.
The principle
The idea is old. Jerome Saltzer and Michael Schroeder set it out in 1975, in a paper on protecting information in computer systems: every program and every user should operate with the smallest set of privileges needed to do the job. It has been a basic rule of security design ever since.
Today's standard wording comes from NIST, the US standards body. Its control catalog, SP 800-53, has a control called AC-6, Least Privilege, which asks organizations to allow users, and processes acting for them, only the authorized access "necessary to accomplish assigned organizational tasks." NIST's glossary gives the same idea in a few versions. All of them have two parts. The access is minimum: not "what might be handy one day", but what the task needs. And it is current: tied to the work someone does now. Access that made sense two jobs ago is no longer least privilege.
Three related ideas often get mixed up with it:
| Idea | What it means | How it relates |
|---|---|---|
| Need to know | People see only the information their work requires | Least privilege applied to data rather than actions |
| Separation of duties | No single person can complete a risky process alone (pay a supplier they added) | Splits privileges so that no one holds the whole chain |
| Role-based access control | Access is given through roles, not person by person | A way to implement least privilege at scale |
Least privilege is the goal. Roles, reviews and admin policies are how you reach it.
Why it matters more than it sounds
A privilege does damage in three ways, and none needs bad intent.
Stolen accounts. When someone's password is phished, the attacker gets exactly what that person had. A phished member account in your CRM exposes their deals. A phished CRM admin can export every customer and add their own user.
Mistakes. Admins can delete workspaces, change billing, turn off logging and remove other admins. The fewer people who can, the fewer accidents.
Leavers. Every account that survives offboarding keeps whatever it had. A forgotten member account is a problem; a forgotten owner account is a door. Orphaned accounts covers how they pile up.
Limiting privileges doesn't stop any of these. It caps the damage.
Why small companies drift from it
Nobody decides to give everyone admin. It happens on its own, for reasons that all make sense at the time.
The first user is the owner. Whoever signs up for a tool becomes its owner or admin by default. In a young company that is often the founder, for every tool, forever.
"Just make me admin." Someone needs to add an integration or invite a guest, and the quickest fix is a role upgrade. Nobody downgrades them afterwards.
People move. The engineer who ran the website for a quarter keeps the marketing tools. The salesperson who helped with support keeps the help desk. This is called privilege creep: access accumulates with every role change because adding is urgent and removing is not. The JML process treats these moves as an event of their own for exactly this reason.
Cheaper plans have cruder roles. On some tools' entry plans the role choice is effectively admin or member, and finer-grained roles (billing-only, read-only, custom roles) arrive on higher plans. GitHub, for example, reserves custom organization and repository roles for Enterprise Cloud, per its documentation as we read it on 27 October 2026. Check your own tools' plan pages; it varies.
Shared logins. A password everyone knows carries the full rights of that account for everyone, with no record of who did what. Shared accounts covers the fix.
Integrations. A person connects an app with "Sign in with Google" or an API token and grants it broad scopes. The app now holds access that nobody reviews and that does not leave when the person does.
The result, if nobody has looked in a while, is usually more admins than anyone would guess and member lists that nobody could explain line by line.
Admin rights
Start here, because this is where least privilege pays off fastest.
Who should have admin rights? For each tool, two named people: a primary and a backup, so you are never locked out and never need to share a login. For your identity system (Google Workspace, Microsoft 365) the same rule applies to super admins. Google's own best practices for administrator accounts recommend more than one super admin, each managed by a separate person, and using limited admin roles for daily admin work.
Separate admin work from daily work. Google advises against using a super admin account for daily activities. NIST's AC-6 has an enhancement, AC-6(2), with the same idea: people with privileged access use a non-privileged account for everything else. In a small company the practical version is a separate admin account for your directory and, in other tools, not doing everyday work while signed in as the owner where you can avoid it.
Prefer the narrowest admin role. Many tools split admin into parts: billing, user management, security settings, content. The person who pays invoices needs billing access, not the power to export all data. Google Workspace, for example, has prebuilt admin roles such as User Management Admin and Help Desk Admin alongside Super Admin.
Grant admin for a task, then take it back. If someone needs admin to set up an integration, give it, note it, and remove it when they are done. Large companies automate this as "just-in-time" privileged access. A calendar reminder works too.
Write the admin list down. One line per tool: who the admins are, and why. If you cannot justify a name, it should not be there.
How to reduce admin access in one afternoon
Begin by listing, for each tool, everyone with an owner, admin or equivalent role. Then mark the two who should keep it, preferring people whose job involves the tool. Downgrade everyone else to the highest non-admin role they need, and tell them first, so the change is a decision they heard about rather than a broken workflow they discover.
Next, remove admin from anyone who has left or is suspended in your directory. This is often the biggest finding. Finally, check that each tool's billing contact and owner account belong to a current person, not a leaver's email; an owner account tied to a departed founder's address is a common and awkward discovery.
Roles inside tools
Each tool has its own role names, but most follow the same tiers:
| Tier | Typical names | Who should hold it |
|---|---|---|
| Owner | Owner, primary owner, account owner | One or two people, usually a founder or ops lead |
| Admin | Admin, workspace admin, manager | The tool's named admins |
| Full member | Member, user, editor, standard | People who use the tool for their job |
| Limited member | Viewer, commenter, read-only, light | People who need to see, not change |
| Guest | Guest, collaborator, external | Contractors, clients and freelancers, scoped to what they work on |
Least privilege inside a tool means picking the lowest tier that lets the person do their work, and using the tool's own scoping where it exists: a guest limited to one project, a member limited to one team's space, a viewer in the finance tool rather than an editor.
Guests are where least privilege and cost line up. In Slack, a single-channel guest can reach one channel and is free, up to five per paid active member, while a multi-channel guest is billed as a regular member, according to Slack's guest help article as we read it on 27 October 2026. The narrowest role is also the cheapest. Contractor access management covers the rest of the contractor case.
Do the same for anything that is not a person. API tokens and app integrations should get the narrowest scopes the tool offers, belong to a role or team account rather than one employee where the tool allows it, and appear on the same list as people. In Google Workspace, the API controls page shows which third-party apps your users have connected and which Google services each one requests.
Is least privilege practical in a startup?
Yes, if you aim it at the right things. Nobody needs a ticket to get a viewer seat in the design tool. What matters is that owner and admin roles are rare, that people who change jobs lose what they no longer need, and that leavers lose everything. Roles help, because you decide once what an "Engineer" or "Sales" person gets, and then apply it to each new person without thinking.
The failure mode to avoid is the opposite one: locking things down so tightly that people share logins or route around you to get work done. A shared password is a worse outcome than a slightly generous role. Least privilege that people actually follow beats a stricter version they work around.
Reviews
Least privilege is a state you have to keep, not one you reach once. NIST's AC-6(7), Review of User Privileges, asks for exactly this: review, at a frequency you define, the privileges assigned to roles or users, and remove what is no longer needed.
For a small company, that means four habits. Quarterly, check every admin list for the tools that hold customer data, money or code; it takes minutes per tool. At every role change, remove the old role's access, which is the step that stops privilege creep. Twice a year, run a full user access review of everyone in every tool, and have the results signed off if you need evidence (access certification explains how). And after any incident, review first: if an account was compromised, check what else that person or token could reach.
The findings should feed back into your defaults. If every review finds marketers with admin in the website builder, change the default role in the marketing bundle. And write the rules down once, in an access control policy, so the next admin inherits them instead of reinventing them.
Where SmashSSO fits
SmashSSO helps with the parts of least privilege that are about people and roles across many tools. It reads your people from Google Workspace with read-only access and shows, in one access grid, who has an account in each tool and with which role, matched by email. Admins stand out, accounts that match nobody in your directory are flagged, and people suspended in Google Workspace show where they still have access.
Access bundles set the role each job gets in each tool, such as "Engineer" as a member in GitHub and a viewer in the design tool, so new people start at the least they need. Role changes and removals run the same way, through a tool's member API where you have added a token or an agent in your own Chrome otherwise, which needs that browser open and can fail when a tool redesigns its screens. Steps that would add a paid seat wait for approval, and everything goes into an audit log. It is not SSO and does not control what people do inside a tool. It costs a flat fee from $199 a month, billed annually (pricing).