Welcome to the M365 FM podcast, where we explore the people, ideas and technologies shaping the future of work. What row level security actually does? Imagine opening a power BI report and seeing numbers that were never meant for you. Nobody hacked anything because the report itself handed you the wrong slice of data. You'd assume the fix is tighter permissions, but the real answer is a setting that's easy to overlook until something goes wrong. It's called row level security, and it works less like a locked door and more like a pair of glasses. Welcome to the M365 FM podcast, where we explore the people, ideas and technologies shaping the future of work. By the end of this, you'll know how power BI decides what each person sees and the places where that protection stops working. The problem shows up the moment one report has to serve a crowd. Sales wants their own numbers, finance wants the totals, and a regional manager should only see the region they actually run, handing everyone the same unfiltered view as how private figures end up on the wrong screen. Row level security, often shortened to less, is how power BI solves that inside a single report. Instead of building a separate copy for every audience, you keep one model and let it decide row by row what each person is allowed to see. It helps to picture those glasses. Again, everybody opens the same report and the glasses decide which rows make it onto the page. The charts, the buttons, and the layouts stay identical while the data underneath them changes from person to person. Three simple pieces make it work. A role is a name for a group of people who should see the same thing. A filter is the rule that defines that thing, like showing only rows where the region is west. Membership is the list of people you place inside that role. So a role called West Sales holds one rule. And the rule says the region must be west. When you add someone from that region to the role, power BI keeps every other region out of their view. They still open the exact report that everyone else opens. That last point is worth repeating, because beginners usually expect the opposite. LSS doesn't hide columns, and it doesn't lock anyone out of the report. It filters rows, and everything else about the report stays exactly where it was. How power BI decides what you see. When you write a rule, you're really writing a test that power BI runs against every single row. For each row, the test answers one simple question, and the answer is either true or false. If it's true, the row appears. If it's false, the row is gone, removed completely rather than grayed out or tucked behind a click. That detail matters more than it sounds, because a filtered row isn't sitting there waiting to be revealed. It simply isn't part of the data power BI hands to that person. Let me make it concrete. Picture a sales table with five rows where two are west, two are east and one is south. A rule reading region equals west turns those five rows into two. For anyone inside that role, the east and south rows never reached their screen, so the totals, the charts, and the drill downs all add up to a smaller and safer picture. You build all of this in power BI desktop. Before you share anything with anyone, open the model, create a role, and attach a rule to it. And when you publish those roles, travel with the model. The security is already in place the moment it lands in the service. Now here is where it gets interesting. Power BI has two very different ideas that both use the word role, and mixing them up causes real trouble. The role you build for security lives inside the model. The role someone holds in a workspace is about who can edit things, and that second one overrides your rules. Anyone with the admin member or contributor role in a workspace can see every row, no matter what filters you wrote, only people holding the viewer role get filtered. So the moment you hand a manager contributor access, you've handed them the whole table. That single detail explains a lot of confusing conversations. Someone builds careful rules, shares the report, and then watches a colleague browse straight past them. The rules aren't broken. The colleague simply sits in a workspace role that sits above them. The same logic explains the blank report. If someone has no role assigned at all, power BI still enforces security, but there's no rule that matches them, so they see nothing. A blank screen usually means a missing or mismatched assignment rather than a broken report. There's one more quirk worth knowing when a row is filtered out. Power BI shows the same kind of message it shows for a field that was deleted or never existed. To the person staring at it, the report looks broken, and that's a rough way to discover your access has changed. The good news is that you don't have to write a fresh rule for every new person who joins the team. There's a way to build a rule that updates itself, and once you see it, you won't want to go back setting it up without writing code. If the phrase security rules makes you picture code. Relax, because most of this is clicking power BI desktop has a point and click editor for the common cases, and you only reach for a formula when you want the fancier behavior. The path looks like this. Open your model, find manage roles on the ribbon and create a new role. Give it a name that means something like West sales because a week from now you'll be glad you did. Then pick the table you want to protect and write the rule with the drop down editor. Save it and the role exists. One thing catches people out here and it deserves a plain answer. You can't add people to a role inside desktop because that part of the job belongs to the service. Desktop is where the rules are written, and the service is where people are placed into them. Next comes the part beginners often skip. Publish the report to the power BI service. The roles ride along with the model. So the rules you built are now live in the cloud. The final step happens in the service. Open the dataset, find the security page, and add people or groups to the role you created. You can type a person's email address directly, or you can point at a group and manage membership in one place. From then on. That last option deserves a closer look because not every group type works here distribution groups, mail enabled groups and Microsoft Android security groups all work fine. Microsoft three hundred sixty five groups don't, which surprises almost everyone the first time they try it. Before you trust any of this, test it in desktop use view as to see the reports through a. Rolls eyes in the service, open the roll and use test as roll to check what that person would see. Just be careful about trusting that test too far. Tesla's roll doesn't work on models that connect live with single sign on, and it always runs as you, so it can't fully show you what another person sees. When a report really matters, sign in as a real test account and look at it yourself. And there's two kinds of rules you can build back to that Self-updating rule rules come in two flavors, and the difference is where the value comes from. In the simple flavor, you hard code the value. So the rule says the region must equal west and it never changes. Build one of those for each audience. Add the right people and you're done. It's fast, it's easy to read, and it's a fine choice when you have a few fixed groups. The catch is maintenance. Every new region, every reorganization, and every new team means another roll to build and another list of people to keep current. Do that for one hundred salespeople and the work starts to own you. The second flavor fixes that by asking who is signed in. Instead of hard coding west, the rule looks up the signed in person's email in a small table that maps each person to their region. That table is your security list, and it can live in a spreadsheet or a database wherever you already keep that kind of thing. Now, the payoff when a new salesperson joins you, add one row to that table and the rule filters their data correctly the moment they sign in. No new role, no new assignment, just one row. That's the version most teams end up using because it grows without turning into a chore. To make it work, power BI needs to know who is looking. It reads the signed in identity, which usually looks like an email address and compares it to the value in your table. If those two don't match exactly, the person sees nothing. So a stray space or an old address can cause a blank report. There's a smaller trap hiding in that identity, and it catches people during testing. The value power BI reads can come back in a slightly different shape depending on whether you're checking in desktop or viewing the report in the service. A rule that works perfectly on your laptop can fail for everyone else, so compare the exact value your users actually get before you trust the rule. One more thing to know before you build rolls add up and that surprises people. If someone belongs to two rolls, they see the combined rose from both rather than only the narrower one. The clearest example comes from payroll. Imagine one roll that shows no payroll rose, and another that shows all of them give a single person both rolls and they see everything because the rules are added together rather than weighed against each other. The familiar idea that a denial beats an allowance simply doesn't hold here. So these rules are powerful and a little dangerous at the same time. The best way to trust them is to see where teams already put them to work, where teams actually use it. These rules earn their keep every day because teams lean on them in situations you'd recognize from your own job. Once you know the pattern, you start spotting it everywhere. Take sales as an example. A regional manager opens the shared dashboard and sees only the numbers for their territory, while the head of sales opens the very same dashboard and sees the whole country. Same report, same charts, two very different pictures, and neither person had to ask for a special version. Healthcare leans on this even harder. A clinician needs their own patient's records, and they shouldn't be able to browse the rest of the hospital just because they have a login row level security keeps that boundary inside the one system everyone already uses. Software companies use it to serve many customers from a single product. Each client opens the same app and the same dashboard, and the rules make sure they only ever see their own data. They never see the customer sitting next to them in the same database. The pressure on those companies is real, because one mistake in the mapping table can blur the line between two customers, and that's the kind of error that ends contracts. Then there's the everyday office version, which is often the first place a beginner meets this. A finance team shares one cost report with every department. Each department had seized their own budget line, and only the finance lead sees the total across the company. What this replaces is the old habit of copying before rules like these. Keeping people apart usually meant building a second report, then a third, then a folder full of near-identical versions. Each copy was another thing to update, and another place for a mistake to hide. One filtered model removes most of that busy work, and it removes the moment where the wrong copy gets handed to the wrong group. The reason all of this works so cleanly comes down to how the filter behaves because it removes rows at the source. A slicer can't reveal them, and a drill down can't dig them back up. If a row never reaches someone, it's gone from the totals, from every chart and from each detail view built on top of it. That sounds airtight, and for the data itself, it is. The trouble rarely comes from the rules, and it almost always comes from the ordinary steps around them. The mistakes that catch beginners. Most of the pain around row level security traces back to a short list of mistakes, and every one of them looks like a bug until you learn the cause. The first is the one we already met, and it's the most common of all. Someone gets a workspace role like admin, member or contributor instead of viewer, and they see everything. If a report leaks and your rules look perfect, check that workspace role before you touch anything else. The second mistake is forgetting to publish again. Rules live in the model, so a change you make in desktop stays on your machine until you push it up. Meanwhile, the service keeps serving the old version and you spend an hour debugging a rule that was never really there. The third one hides in plain sight. Your rule compares the signed in identity to a value in your table. And if they don't line up, the person sees a blank report, a nickname, an old address, or a trailing space is enough to break the match. When someone reports an empty screen, compare their exact sign in value to the table before you rebuild anything. The fourth mistake is building one role for every single person. It works at first, and then the team grows and every new hire turns into a small manual project. Groups and a mapping table exist, so you never have to live that way. The fifth mistake is assuming the rules protect more than they do. They protect the rows in the model and nothing beyond it. They don't protect the source file sitting on someone's laptop, and they don't protect a spreadsheet. You export and hand around by mistake. There's a mirror image of the fourth mistake, and it shows up much later. A team builds that mapping table, fills it in carefully, and then never looks at it again. People change roles and leave the company, and older dresses sit in the table long after the person has gone. Keep the list alive or the security drifts out of step with reality. The sixth is skipping the real test view as in test as role are handy, but they run as you so they can hide the exact mismatch you are hunting for. A quick sign in as an actual test user catches problems. The built in tools never show. Those tools still have a place, and they're fast enough for a first pass, so keep using them. Just don't let them be your only proof that the security works, because the gap they leave is exactly where the real problem likes to hide. None of these mistakes are exotic. They're the ordinary slips that come from doing everything right except the very last step. What row level security can't do. Now for the honest part. Because knowing where a tool stops working is what separates a calm deployment from a nasty surprise. Row level security filters, rows. And that's the whole job. It can't hide a column. If someone can see a row, they can see every field inside it, including the salary or the margin. You'd rather keep quiet. Hiding columns is a separate feature called object Level security, so don't confuse the two. It also can't show a safe summary while hiding the detail underneath. If a person can reach the total, they can reach the rows that built it, so you can't hand someone a friendly number while keeping the receipts locked away. The filter only travels along active relationships in your model. If a table isn't connected to the rule, it isn't protected, which is why a well-built model matters as much as the rules themselves. Get the model right and the filter flows where you expect it to. It's also worth knowing where these rules don't apply at all. When power BI connects live to a model that lives somewhere else, such as a hosted analysis services model that outside model handles its own security. There's nothing to set up here, and the security option won't even appear, which confuses people who go looking for it. There's a performance cost to every query. Now carries that extra filtering so a heavy report can slow down. You can measure the difference by comparing query times with the rules on and off. It's rarely dramatic, but it's real, so check it before you scale up. Some doors stay open by design workspace admins, members, and contributors always see everything, and a report reached through an automated service account isn't filtered either. Features like publish to web stripped the security out entirely, so never point them at protected data. And sometimes the honest answer is to skip it. If your rules are few and fixed, publishing a separate model for each audience is simpler and often faster. As long as you accept more reports to maintain. RL's earns its place when one model has to serve many different eyes. Where this fits in Microsoft three hundred sixty five and fabric step back and this stops looking like a power BI trick and starts looking like one layer in a bigger system. Row level security sits beside object level security, which hides columns and tables and workspace permissions which decide who gets through the door at all. Together, they shape how your data is protected in Microsoft Fabric. The same idea carries straight over. It works on fabric semantic models, including the newer direct lake kind. So what you learn here still applies when your data moves into fabric. And because the rules lean on the same Microsoft Entra sign in that runs teams, SharePoint and the rest of Microsoft three hundred sixty five. One identity follows your people across every surface. So here's the short version. Row level security is the same report showing a different slice to each person, and it filters rows through roles you build in desktop and assign in the service. The real tension here is never the setup, and it's remembering who your rules leave out. That leads me to a question for you. Which report in your world would you want to protect first? And who is currently seeing more of it than they should? If this made the topic feel less intimidating, subscribe to M365 FM to master the tools and the mindset shaping the future of work.