WEBVTT 00:00:03.660 --> 00:00:06.240 Welcome to the Azure Security Podcast, where 00:00:06.240 --> 00:00:08.740 we discuss topics relating to security, privacy, 00:00:09.060 --> 00:00:11.460 reliability, and compliance on the Microsoft 00:00:11.460 --> 00:00:16.219 Cloud Platform. Hey, everybody. Welcome to Episode 00:00:16.219 --> 00:00:19.539 110. This week is myself, Michael, and I'm here 00:00:19.539 --> 00:00:22.280 with Gladys, and our guest is Sharon, who's here 00:00:22.280 --> 00:00:26.289 to talk to us about... Securing Gen AI Apps with 00:00:26.289 --> 00:00:28.989 Antro. This is number three of four. So if you 00:00:28.989 --> 00:00:31.469 haven't seen or listened to the other two, I 00:00:31.469 --> 00:00:33.630 would probably urge you. They are kind of independent 00:00:33.630 --> 00:00:35.350 of each other, but it's worthwhile looking at 00:00:35.350 --> 00:00:38.409 episodes one and two. But before we get to our 00:00:38.409 --> 00:00:40.869 guest, let's take a little lap around the news. 00:00:41.189 --> 00:00:43.469 Gladys, why don't you kick things off? Okay, 00:00:43.509 --> 00:00:47.429 so I only have two news articles that I'm going 00:00:47.429 --> 00:00:51.070 to talk about. The first one is regarding... 00:00:52.520 --> 00:00:56.880 Defender for Endpoint in Azure Government. As 00:00:56.880 --> 00:01:02.079 some of you may know, we are deprecating Log 00:01:02.079 --> 00:01:07.420 Analytics Agent, or MMA, if you see that acronym 00:01:07.420 --> 00:01:14.439 around, in Azure. So in order to enable capabilities 00:01:14.439 --> 00:01:21.040 with file integrity, a new file integrity monitoring, 00:01:21.340 --> 00:01:25.489 or FIM, has been developed and now is general 00:01:25.489 --> 00:01:29.930 availability in Azure Government. So just make 00:01:29.930 --> 00:01:35.629 sure that you are able to migrate this configuration 00:01:35.629 --> 00:01:39.810 because soon that functionality will stop working. 00:01:40.049 --> 00:01:42.569 But basically, it allows you to meet compliance 00:01:42.569 --> 00:01:45.629 requirements by monitoring critical files and 00:01:45.629 --> 00:01:49.549 registering in real time and auditing the changes. 00:01:51.519 --> 00:01:55.359 It also helps you identify potential securities 00:01:55.359 --> 00:02:01.219 issues by detecting suspicious files, file content 00:02:01.219 --> 00:02:04.920 changes, and things like that. The next news 00:02:04.920 --> 00:02:07.180 that I wanted to talk about, which I think is 00:02:07.180 --> 00:02:10.979 very relevant to this talk today, so I'm not 00:02:10.979 --> 00:02:16.020 going to talk a lot about it, is about Microsoft 00:02:16.020 --> 00:02:20.639 Graph. Microsoft Graph is one of the most important 00:02:20.639 --> 00:02:26.919 aspect that enables AI. And in order to secure 00:02:26.919 --> 00:02:30.659 access that is being done through the Microsoft 00:02:30.659 --> 00:02:33.520 Graph, there's several permissions that have 00:02:33.520 --> 00:02:36.780 been developed and more extensions are being 00:02:36.780 --> 00:02:40.080 enabled. So there's more extensions that have 00:02:40.080 --> 00:02:43.330 been released lately. There's an article that 00:02:43.330 --> 00:02:46.770 I'm putting as part of the website. And hopefully 00:02:46.770 --> 00:02:50.449 we talk a little bit more during our talk in 00:02:50.449 --> 00:02:52.689 the next few minutes. I have a couple of items. 00:02:52.930 --> 00:02:54.370 Actually, I have more than a couple. I have like 00:02:54.370 --> 00:02:57.710 five or six items. So the first one is we've 00:02:57.710 --> 00:03:00.860 now... added a thing called API Security Posture 00:03:00.860 --> 00:03:03.719 Management. As you probably know, most cloud 00:03:03.719 --> 00:03:06.280 -based applications use REST endpoints of some 00:03:06.280 --> 00:03:09.800 kind. And what this new capability allows is 00:03:09.800 --> 00:03:12.080 things like identifying APIs that are exposed 00:03:12.080 --> 00:03:14.639 to the internet, automatically discovering sensitive 00:03:14.639 --> 00:03:16.340 information that may be traveling through those 00:03:16.340 --> 00:03:18.900 endpoints. There's all sorts of other things 00:03:18.900 --> 00:03:20.860 that it does as well. So well worth looking into 00:03:20.860 --> 00:03:23.479 that if you're using REST endpoints, which odds 00:03:23.479 --> 00:03:26.650 are good you are. Next item is the general availability 00:03:26.650 --> 00:03:30.250 of FEP 140 Level 3 HSM's premium key vault is 00:03:30.250 --> 00:03:32.830 now available in China. This is good to see. 00:03:32.849 --> 00:03:34.789 It's good to see, you know, very robust ways 00:03:34.789 --> 00:03:36.349 of storing credentials and secrets and keys. 00:03:37.189 --> 00:03:39.409 Next one, which has got nothing to do with securing 00:03:39.409 --> 00:03:43.150 or storing credentials at all, which is the way 00:03:43.150 --> 00:03:46.069 you should go, which is Fabric Data Warehouse 00:03:46.069 --> 00:03:49.780 now supports service principles. huge fan of 00:03:49.780 --> 00:03:51.819 using things like service principles and managed 00:03:51.819 --> 00:03:53.819 identities rather than storing a credential somewhere. 00:03:54.159 --> 00:03:56.759 So this is always good to see. Next one, which 00:03:56.759 --> 00:03:58.979 actually has nothing to do with security whatsoever, 00:03:59.240 --> 00:04:00.539 but the only reason I'm really talking about 00:04:00.539 --> 00:04:02.620 this because A, it's important, and B, it's my 00:04:02.620 --> 00:04:04.979 old stomping ground, which is Azure Data. So 00:04:04.979 --> 00:04:08.719 Azure Data Studio is being retired. The actual 00:04:08.719 --> 00:04:11.439 retirement will be, I believe, like February 00:04:11.439 --> 00:04:14.939 28th, 2026, so next year. And what we're recommending 00:04:14.939 --> 00:04:18.639 people do is go to use Visual Studio Code with 00:04:18.639 --> 00:04:21.160 the appropriate MS SQL extension. If you kind 00:04:21.160 --> 00:04:22.319 of look at it, I don't know if you guys know 00:04:22.319 --> 00:04:23.800 or not, but essentially they're a very, very 00:04:23.800 --> 00:04:26.600 similar tool. They're written using Electron, 00:04:26.620 --> 00:04:29.199 both tools are. And the code bases, you know, 00:04:29.220 --> 00:04:30.939 it's kind of similar in some way. They kind of 00:04:30.939 --> 00:04:33.720 look the same and feel the same. It didn't really 00:04:33.720 --> 00:04:36.319 make sense having two bits of code that were 00:04:36.319 --> 00:04:39.139 kind of the same. So anyway, that's what we're 00:04:39.139 --> 00:04:40.420 going to do. So as your data studio is being 00:04:40.420 --> 00:04:42.420 retired and you should be moving over to using 00:04:42.420 --> 00:04:45.000 VS Code with the MS SQL extension and frankly, 00:04:45.100 --> 00:04:46.819 all the other extensions that you would normally 00:04:46.819 --> 00:04:49.319 have as well. A good friend of the podcast, Meryl 00:04:49.319 --> 00:04:51.759 Fernando, has a couple of posts up on X talking 00:04:51.759 --> 00:04:54.839 about Microsoft Graph permissions as well as... 00:04:55.500 --> 00:04:57.660 Talking about some ways of detecting device code 00:04:57.660 --> 00:05:01.000 flow. So the first one on Microsoft Graph, there's 00:05:01.000 --> 00:05:02.800 a new least privilege permission that's been 00:05:02.800 --> 00:05:05.199 added. Historically, if you wanted to see if 00:05:05.199 --> 00:05:08.019 somebody had their account enabled or disabled, 00:05:08.379 --> 00:05:10.180 you would probably have to grant something like 00:05:10.180 --> 00:05:13.920 readwrite .all, which is... A lot more than just 00:05:13.920 --> 00:05:15.879 asking to determine if an account is enabled 00:05:15.879 --> 00:05:18.420 or not. So the Microsoft Graph folks have actually 00:05:18.420 --> 00:05:22.160 added a new grant called Enable Disable Account. 00:05:22.540 --> 00:05:25.660 And just grant an account or a principal that 00:05:25.660 --> 00:05:27.939 particular permission and off you go. So you're 00:05:27.939 --> 00:05:29.839 really restricting what the user can actually 00:05:29.839 --> 00:05:33.000 access. And then the last one, the ability to 00:05:33.000 --> 00:05:35.519 block device code flow. And this is available 00:05:35.519 --> 00:05:38.199 now in Microsoft Entry ID Conditional Access. 00:05:40.110 --> 00:05:42.829 device code flow just simply should just not 00:05:42.829 --> 00:05:46.870 be used. It's an incredibly insecure OAuth 2 00:05:46.870 --> 00:05:50.189 flow and you just shouldn't use it. And so now 00:05:50.189 --> 00:05:52.470 you can block it, which is really, really cool. 00:05:52.569 --> 00:05:54.319 All right. So now we've got the news out of the 00:05:54.319 --> 00:05:57.540 way, let's turn our attention to our guest. This 00:05:57.540 --> 00:05:59.220 week, as I mentioned, it's Sharon, and she's 00:05:59.220 --> 00:06:02.420 here to talk to us about securing Gen A apps 00:06:02.420 --> 00:06:05.079 with Entra. Again, this is part three of a four 00:06:05.079 --> 00:06:07.839 -part series, and the focus in this particular 00:06:07.839 --> 00:06:10.959 chat is about monitoring and governance. So Sharon, 00:06:11.000 --> 00:06:13.160 welcome to the podcast. Would you like to take 00:06:13.160 --> 00:06:15.199 a moment and introduce yourself to our listeners? 00:06:16.079 --> 00:06:18.139 Yeah, thanks, Michael. Yeah, my name is Sharon 00:06:18.139 --> 00:06:21.560 Chahal. I'm a principal product manager within 00:06:21.560 --> 00:06:24.259 our customer experience engineering team within 00:06:24.259 --> 00:06:28.259 ID &A. I have the pleasure of leading our efforts 00:06:28.259 --> 00:06:32.100 within Security Copilot and Entra. Identity protection. 00:06:32.379 --> 00:06:34.420 So identity protection is the risk -based policies, 00:06:34.699 --> 00:06:37.259 conditional access policies, and looking at all 00:06:37.259 --> 00:06:40.699 up in securing AI as well. So it's particularly 00:06:40.699 --> 00:06:43.439 initiatives of helping our organizations out 00:06:43.439 --> 00:06:46.199 there to protect their assets, protect their 00:06:46.199 --> 00:06:48.300 identities, and sort of really protect their 00:06:48.300 --> 00:06:50.959 devices when they start exploring the world of 00:06:50.959 --> 00:06:54.319 Gen I. So this is all about monitoring. So can 00:06:54.319 --> 00:06:56.720 you sort of explain what that means and also 00:06:56.720 --> 00:07:00.279 kind of how it relates, if it does, to governance? 00:07:01.389 --> 00:07:04.949 Absolutely. So when we're monitoring access for 00:07:04.949 --> 00:07:08.329 securing a guy, it's ensuring that only authorized 00:07:08.329 --> 00:07:12.449 users and systems can interact with AI resources. 00:07:12.689 --> 00:07:16.269 So, you know, in cases, AI systems may handle 00:07:16.269 --> 00:07:20.230 sensitive data and make critical decisions. So 00:07:20.230 --> 00:07:23.750 controlling who can access these systems, preventing 00:07:23.750 --> 00:07:27.709 misuse or data breaches, or if you think about 00:07:27.709 --> 00:07:31.699 malicious manipulation. By continuously tracking 00:07:31.699 --> 00:07:35.800 access patterns and behaviors, you can identify 00:07:35.800 --> 00:07:38.800 or organizations can identify potential threats. 00:07:39.449 --> 00:07:42.629 or suspicious activities such as that unauthorized 00:07:42.629 --> 00:07:46.490 attempt to influence an ai decision making or 00:07:46.490 --> 00:07:50.129 gain access to confidential information so when 00:07:50.129 --> 00:07:52.649 we really think about it it's the level of monitoring 00:07:52.649 --> 00:07:56.810 that provides a proactive defense ensuring the 00:07:56.810 --> 00:08:00.949 integrity along with the security of both AI 00:08:00.949 --> 00:08:05.009 and its surrounding infrastructure as well. So 00:08:05.009 --> 00:08:07.930 it's really crucial to ensure the reliability 00:08:07.930 --> 00:08:11.209 of those particular governance requirements and 00:08:11.209 --> 00:08:13.410 being able to monitor them to prevent the vulnerabilities 00:08:13.410 --> 00:08:16.889 from being exploited and also ensuring the integrity 00:08:16.889 --> 00:08:20.610 of AI -driven processes and data. So recently 00:08:20.610 --> 00:08:26.709 I was working on an attack chain that I was drawing 00:08:26.709 --> 00:08:33.440 regarding how Malicious actors may use AI. And 00:08:33.440 --> 00:08:38.379 I realized that with AI, of course, the original 00:08:38.379 --> 00:08:42.740 exposure may be like the normal one, emails, 00:08:43.120 --> 00:08:48.240 endpoint, networks, so forth. But at the end, 00:08:48.320 --> 00:08:53.679 where they meet is identity, applications, and 00:08:53.679 --> 00:08:59.000 data. how do we control these let's start maybe 00:08:59.000 --> 00:09:01.860 like access review can you talk a little bit 00:09:01.860 --> 00:09:05.610 about that So one of the key things that we think 00:09:05.610 --> 00:09:07.549 about, and you brought this up around, is around 00:09:07.549 --> 00:09:09.950 access permissions, right? So access permissions 00:09:09.950 --> 00:09:12.970 is really identified as a security risk. So say, 00:09:13.049 --> 00:09:16.429 for example, you have an employee with overly 00:09:16.429 --> 00:09:19.590 broad access that modifies, for example, an AI 00:09:19.590 --> 00:09:22.149 module or some data they're not meant to get 00:09:22.149 --> 00:09:24.870 access to. It could be confidential, like financial 00:09:24.870 --> 00:09:29.330 data. So it's really exposing it to unauthorized 00:09:29.330 --> 00:09:33.279 users. So when we look at Entry ID and helping 00:09:33.279 --> 00:09:36.179 how to sort of monitor and mitigate that, we 00:09:36.179 --> 00:09:38.600 have, you know, RBAC, which is role -based access 00:09:38.600 --> 00:09:41.539 control and least privilege. So that's restricting 00:09:41.539 --> 00:09:45.320 the AI module access to necessary roles that 00:09:45.320 --> 00:09:46.919 do actually need access. And then we talk about 00:09:46.919 --> 00:09:49.340 access reviews. Access reviews is really down 00:09:49.340 --> 00:09:52.659 to regular auditing and removing unnecessary 00:09:52.659 --> 00:09:56.000 permissions. So say if someone in the organization 00:09:56.000 --> 00:10:00.220 had left. You have a process where we've spoken, 00:10:00.419 --> 00:10:02.500 I've spoken to multiple organizations where they 00:10:02.500 --> 00:10:07.360 automatically trigger perhaps a... a flow to 00:10:07.360 --> 00:10:10.139 a manager to say, hey, this person is no longer 00:10:10.139 --> 00:10:12.360 an organization. Can you validate that? It'll 00:10:12.360 --> 00:10:15.320 be either click or a yes or a no. That's kind 00:10:15.320 --> 00:10:17.700 of a one -step prong that you can take into to 00:10:17.700 --> 00:10:21.580 be able to ensure that the necessary or unnecessary 00:10:21.580 --> 00:10:24.019 permissions, so we say in this context, are actually 00:10:24.019 --> 00:10:26.440 removed from that user as well. And it could 00:10:26.440 --> 00:10:28.799 be also a case of, hey, there's been a project 00:10:28.799 --> 00:10:31.600 that a particular person could have been working 00:10:31.600 --> 00:10:34.179 on and now that project is finished. Do they 00:10:34.179 --> 00:10:36.759 still continue? Do they need access to that data 00:10:36.759 --> 00:10:39.279 know okay what we need to do have that access 00:10:39.279 --> 00:10:41.960 review do those regulative audits and what we 00:10:41.960 --> 00:10:44.360 found is that with organizations who adopt that 00:10:44.360 --> 00:10:47.120 they've managed to really reduce their attack 00:10:47.120 --> 00:10:50.659 service and remove the the privilege kind of 00:10:50.659 --> 00:10:53.480 the privilege misuse when it comes down to the 00:10:53.480 --> 00:10:56.200 access permission side of things so talking about 00:10:56.200 --> 00:10:59.220 access reviews what kind of core tools would 00:10:59.220 --> 00:11:02.200 you use in order to manage those That's a great 00:11:02.200 --> 00:11:04.159 question, Gladys. This is definitely a question 00:11:04.159 --> 00:11:06.139 I get asked every day. Well, I'd say partially 00:11:06.139 --> 00:11:08.639 when I go and meet my customers is that, you 00:11:08.639 --> 00:11:10.320 know, when we look at the core tools that's available 00:11:10.320 --> 00:11:13.659 in Entra today, we have sort of conditional access 00:11:13.659 --> 00:11:17.899 policies, applying MFA using risk -based access. 00:11:18.480 --> 00:11:22.080 Then we also think about PIM using that's privileged 00:11:22.080 --> 00:11:25.529 identity management. just -in -time as well, 00:11:25.629 --> 00:11:28.909 just access to securing, for example, a service 00:11:28.909 --> 00:11:32.289 account. Looking at sort of the audit logging 00:11:32.289 --> 00:11:35.110 and sign -in logs, we have those available in 00:11:35.110 --> 00:11:38.610 Entra as well. A lot of sort of when we look 00:11:38.610 --> 00:11:42.289 at the anomaly detection, so we also monitor 00:11:42.289 --> 00:11:45.590 the access and behavior of accounts that could 00:11:45.590 --> 00:11:49.710 be classed as risky as well. So we have a plethora 00:11:49.710 --> 00:11:51.509 of products that are available to be able to 00:11:51.509 --> 00:11:55.440 help just time to cater and help organizations 00:11:55.440 --> 00:11:59.179 to be able to reduce the attack surface. All 00:11:59.179 --> 00:12:00.879 right, let's start at the very top then. Okay, 00:12:00.940 --> 00:12:03.299 so we've got access reviews, RBAC, audit logs, 00:12:03.419 --> 00:12:05.080 monitoring, and conditional access. So why don't 00:12:05.080 --> 00:12:06.720 we start at the very, very top. Do you want to 00:12:06.720 --> 00:12:09.259 give us a brief overview of each of these? So 00:12:09.259 --> 00:12:12.669 let's start with access reviews. access reviews. 00:12:13.049 --> 00:12:16.129 We have this capability available already in 00:12:16.129 --> 00:12:19.370 Entra and obviously through Microsoft Identity 00:12:19.370 --> 00:12:21.470 Governance for those who are familiar with it. 00:12:21.710 --> 00:12:24.470 You can set up, so say for example, be part of 00:12:24.470 --> 00:12:27.070 a joiner, mover, leaver process, also referred 00:12:27.070 --> 00:12:31.539 to as a JML process. is putting in automated 00:12:31.539 --> 00:12:35.360 sort of workflows for those, for people who are 00:12:35.360 --> 00:12:37.340 actually coming on board. So you're doing account 00:12:37.340 --> 00:12:40.440 provisioning, you could do that. You can trigger 00:12:40.440 --> 00:12:43.340 that in a template and add them to a particular 00:12:43.340 --> 00:12:46.259 group. So say if they are engaging, for example, 00:12:46.259 --> 00:12:49.259 on an AI project, you'd have a dedicated security 00:12:49.259 --> 00:12:52.220 group, which you can compass within a workflow. 00:12:52.600 --> 00:12:56.240 And as soon as someone joins, they will go through. 00:12:56.279 --> 00:12:58.480 It's just a click of a button. Everything will 00:12:58.539 --> 00:13:01.659 be sort of the access that the need will be designated 00:13:01.659 --> 00:13:05.629 through that workflow. Similar case for deprovisioning 00:13:05.629 --> 00:13:07.850 as well. We know that can be a bit of a hurdle 00:13:07.850 --> 00:13:11.070 for organizations. Coming in, setting up those 00:13:11.070 --> 00:13:14.029 workflows and templates, putting in that as soon 00:13:14.029 --> 00:13:16.570 as a person leaves or an account is blocked, 00:13:16.909 --> 00:13:20.570 a workflow is then triggered and all of the sort 00:13:20.570 --> 00:13:22.690 of the higher privilege access and the access 00:13:22.690 --> 00:13:26.789 that they have to sort of organizational artifacts 00:13:26.789 --> 00:13:30.090 and materials, their systems are completely blocked. 00:13:30.330 --> 00:13:32.610 So it's kind of a one way of what you can do 00:13:32.610 --> 00:13:36.850 around access. Okay, so let's talk a little bit 00:13:36.850 --> 00:13:40.269 about RBAC. What would you do in order to protect 00:13:40.269 --> 00:13:44.120 the data and applications? It's a really good 00:13:44.120 --> 00:13:47.740 question, Gladys. When we think of RBAC, we know 00:13:47.740 --> 00:13:50.139 we're kind of aligning to least privilege, right? 00:13:50.200 --> 00:13:54.460 So in terms of figuring out who your admins are, 00:13:54.620 --> 00:13:56.960 so, you know, an independent, you'd have a couple 00:13:56.960 --> 00:14:00.559 of your global admins, then you'd have sort of 00:14:00.559 --> 00:14:02.879 your application admins, and then you'd have 00:14:02.879 --> 00:14:05.940 your security admins. So these are roles that 00:14:05.940 --> 00:14:09.720 we fundamentally have built in today within Entra, 00:14:09.799 --> 00:14:12.259 which you can also see in the Azure portal as 00:14:12.259 --> 00:14:16.990 well. By really restricting access to, you know, 00:14:17.009 --> 00:14:19.309 as those who require it, the least privileged 00:14:19.309 --> 00:14:22.690 kind of side of the business, you're able to 00:14:22.690 --> 00:14:28.210 really provide, prevent sort of privilege creep. 00:14:28.529 --> 00:14:30.950 So here you're really ensuring that compliance 00:14:30.950 --> 00:14:34.509 with policies that you may have already defined 00:14:34.509 --> 00:14:38.980 within your organization are being met. So by 00:14:38.980 --> 00:14:41.460 allowing, and you could time gate it as well. 00:14:41.580 --> 00:14:44.899 So by saying, look, I'm a developer. I need access 00:14:44.899 --> 00:14:48.740 to this resource by, for example, navigating 00:14:48.740 --> 00:14:52.840 to go to a PIM portal or you could apply it for 00:14:52.840 --> 00:14:55.679 a security group. You're then applying the correct 00:14:55.679 --> 00:14:57.820 permission for that developer. So you're not 00:14:57.820 --> 00:15:01.080 providing an overprivilege. on their account 00:15:01.080 --> 00:15:04.240 so they can run amok, so to speak. You're making 00:15:04.240 --> 00:15:07.639 sure that you are specifically using those RBAC 00:15:07.639 --> 00:15:11.080 roles to protect your overall environment. So 00:15:11.080 --> 00:15:14.909 what tools would you use to monitor that? So 00:15:14.909 --> 00:15:17.769 as I said, you can have that. So they are inbuilt 00:15:17.769 --> 00:15:21.409 in Entra today. You can navigate into the portal. 00:15:21.789 --> 00:15:24.429 Typically what we see is a lot of organizations, 00:15:24.509 --> 00:15:28.009 they create security groups. By going into the 00:15:28.009 --> 00:15:29.970 Entra portal, you can use it. You can create 00:15:29.970 --> 00:15:33.529 a cloud security group, or if in some cases organizations 00:15:33.529 --> 00:15:37.549 are in a hybrid mode, you can also sync your 00:15:37.549 --> 00:15:40.490 groups that you may have on -prem over to the 00:15:40.490 --> 00:15:43.250 cloud. It could be through Entra Connect or Entra 00:15:43.250 --> 00:15:46.320 Cloud Sync. And then using those particular groups 00:15:46.320 --> 00:15:50.840 and targeting it in the in -built roles that 00:15:50.840 --> 00:15:54.519 we have in Entra today. So it is initially applying 00:15:54.519 --> 00:15:56.899 it through there. We don't really see sort of 00:15:56.899 --> 00:15:58.759 scenarios where you would do it individually. 00:15:58.919 --> 00:16:01.279 So say, for example, Sharon needs access to this. 00:16:01.399 --> 00:16:04.100 I'm actually going to apply just the user administrator 00:16:04.100 --> 00:16:07.799 role to her account. It's more, hey, Sharon needs 00:16:07.799 --> 00:16:10.970 access. as a user administrator in this portal, 00:16:11.269 --> 00:16:13.889 or even this particular piece where she might 00:16:13.889 --> 00:16:16.370 be working on a project, we're just going to 00:16:16.370 --> 00:16:18.690 add her to that security group. So that would 00:16:18.690 --> 00:16:20.649 be really consequential of what you would do 00:16:20.649 --> 00:16:23.649 in the portal today. So when you're using PIM, 00:16:23.710 --> 00:16:26.570 Privileged Identity Management, I can time gate 00:16:26.570 --> 00:16:28.169 that as well, right? So Sharon has access for 00:16:28.169 --> 00:16:30.870 the next 60 minutes. Yes, you can absolutely 00:16:30.870 --> 00:16:33.370 do that. Yeah, you can go in, you can customize 00:16:33.370 --> 00:16:35.990 it. So, hey, Sharon's just going to get 60 minutes 00:16:35.990 --> 00:16:37.870 of this because that's all she needs in order 00:16:37.870 --> 00:16:41.309 to complete her task within her particular role 00:16:41.309 --> 00:16:43.690 in this particular project. And then that access 00:16:43.690 --> 00:16:45.809 is completely removed. Okay. And I really want 00:16:45.809 --> 00:16:47.549 to stress just how important lease privilege 00:16:47.549 --> 00:16:50.269 is. When I look at, you know, attacks against 00:16:50.269 --> 00:16:52.490 Azure, like what people are trying to do, one 00:16:52.490 --> 00:16:54.809 of the biggest weaknesses that we see, you know, 00:16:54.809 --> 00:16:57.029 people make is they run applications with much 00:16:57.029 --> 00:16:58.919 higher privilege than is needed. And if that 00:16:58.919 --> 00:17:01.879 application is compromised, you know, the attacker 00:17:01.879 --> 00:17:03.919 is now running with those, you know, elevated 00:17:03.919 --> 00:17:06.259 capabilities. So I really want to stress just 00:17:06.259 --> 00:17:08.619 how critically important lease privilege is. 00:17:08.680 --> 00:17:10.759 And it's certainly a big focus that we have on 00:17:10.759 --> 00:17:14.160 Secure Future Initiative. So just sort of piggybacking 00:17:14.160 --> 00:17:17.500 on what Gladys said. So what kind of audit logs 00:17:17.500 --> 00:17:20.559 are produced and, you know, how can they be consumed? 00:17:21.099 --> 00:17:25.000 Good question there. So within Entry ID, we have 00:17:25.000 --> 00:17:28.579 the sign -in logs. We also have the audit logs. 00:17:28.700 --> 00:17:29.900 You're like, hey, Sharon, what's the difference, 00:17:30.059 --> 00:17:33.680 right? So the activity logs for the audit logs 00:17:33.680 --> 00:17:35.559 is really around sign -in, is around information 00:17:35.559 --> 00:17:39.180 about sign -ins and how your resources are actually 00:17:39.180 --> 00:17:41.279 used by your users within your organization. 00:17:42.089 --> 00:17:44.609 They could also be around provisioning, you know, 00:17:44.650 --> 00:17:49.250 activities that is performed within a provisioning 00:17:49.250 --> 00:17:51.930 service. Like a really good example is a creation 00:17:51.930 --> 00:17:55.190 of a group in ServiceNow. If you create that, 00:17:55.329 --> 00:17:58.809 you're also monitoring that as well. We've been 00:17:58.809 --> 00:18:01.789 able to add even more. It could be the case of 00:18:01.789 --> 00:18:05.109 what groups have been recently added, what applications 00:18:05.109 --> 00:18:07.470 were recently added. I think this seems to be 00:18:07.470 --> 00:18:10.450 a bit of a pain point for our customers is really 00:18:10.450 --> 00:18:12.339 to be able to determine and like, hey, okay, 00:18:12.440 --> 00:18:15.299 we have a ton of applications. How are we able 00:18:15.299 --> 00:18:17.759 to monitor this? You have this in audit logs 00:18:17.759 --> 00:18:22.519 today. Those activity logs, they seem to be that 00:18:22.519 --> 00:18:24.440 central point that people are navigating to. 00:18:24.720 --> 00:18:27.180 And the way they're actually extracting that, 00:18:27.319 --> 00:18:30.660 it could be to a SIEM tool. That's kind of sort 00:18:30.660 --> 00:18:33.180 of the heavy usage, especially for SOC teams 00:18:33.180 --> 00:18:37.529 who need to investigate incidents. So if it could 00:18:37.529 --> 00:18:39.869 be like Sentinel, we know we have Microsoft Sentinel. 00:18:39.930 --> 00:18:42.650 There's a connector there. You can connect that, 00:18:42.670 --> 00:18:46.329 the audit log, straight to an Azure log workspace. 00:18:46.750 --> 00:18:48.970 And from the Azure log workspace, you can connect 00:18:48.970 --> 00:18:51.750 that to Sentinel. So all of that information 00:18:51.750 --> 00:18:55.450 is being correlated in one area for a deep investigation 00:18:55.450 --> 00:18:58.980 that a SOC analyst may need to do. Auditing logs 00:18:58.980 --> 00:19:02.680 help out a lot with security, but a lot of people 00:19:02.680 --> 00:19:06.319 kind of forget that it helps also in planning 00:19:06.319 --> 00:19:09.019 because if you have incorrect configuration, 00:19:09.380 --> 00:19:12.660 you could see it in there. But I love more things 00:19:12.660 --> 00:19:16.059 that can be done in almost real time. So can 00:19:16.059 --> 00:19:18.759 you talk about conditional access as a method 00:19:18.759 --> 00:19:22.470 of protecting in almost real time? Absolutely, 00:19:22.690 --> 00:19:24.930 Gladys. And this kind of brings me to a great 00:19:24.930 --> 00:19:28.869 point. So one of the areas that we've been kind 00:19:28.869 --> 00:19:31.049 of working around in conditional access is really 00:19:31.049 --> 00:19:34.210 to protect AI. So, you know, it could be when 00:19:34.210 --> 00:19:36.990 we say protect AI, we're talking about sort of 00:19:36.990 --> 00:19:40.470 gen AI apps. In this scenario, we could use like 00:19:40.470 --> 00:19:44.589 Microsoft Security Copilot or Microsoft 365 Copilot, 00:19:44.730 --> 00:19:48.549 which are now, you know, organizations are adopting 00:19:48.549 --> 00:19:52.549 these and protecting those services. from misuse. 00:19:53.150 --> 00:19:55.250 can be accomplished. Let's say, for example, 00:19:55.250 --> 00:19:57.549 if you already have conditional access policies, 00:19:58.130 --> 00:20:00.769 which you would set up through the Entra portal 00:20:00.769 --> 00:20:05.130 as well, you can apply those policies to those 00:20:05.130 --> 00:20:09.549 specific Gen AI services, and then that will 00:20:09.549 --> 00:20:11.990 target all of your resources. So that goes down 00:20:11.990 --> 00:20:15.769 to all of your users. We have risky users, which 00:20:15.769 --> 00:20:17.650 is part of conditional access policies when we 00:20:17.650 --> 00:20:23.150 think of risk -based users. You can and users 00:20:23.150 --> 00:20:27.390 with insider risk as well. And so what we found 00:20:27.390 --> 00:20:30.240 is through conditional access policies. Target 00:20:30.240 --> 00:20:32.460 your application, right? One of the big things 00:20:32.460 --> 00:20:35.700 that we've definitely been speaking to across 00:20:35.700 --> 00:20:38.420 the board is complete vision -resistant MFA, 00:20:38.619 --> 00:20:40.880 right? When we think about it, vision -resistant 00:20:40.880 --> 00:20:44.460 MFA from pass keys to Windows Hello for Business 00:20:44.460 --> 00:20:47.980 or even a point of using a FIDO2 key, for example, 00:20:48.099 --> 00:20:50.880 that's a really good way to be able to secure 00:20:50.880 --> 00:20:55.380 who's logging in. So really that's the authentication 00:20:55.380 --> 00:20:58.210 piece. And once you pass that authentication 00:20:58.210 --> 00:21:01.089 piece, it's an authorization, right? So you can 00:21:01.089 --> 00:21:05.910 apply compliant devices. So with conjunction 00:21:05.910 --> 00:21:09.230 with Intune, if you're using Intune, you use 00:21:09.230 --> 00:21:12.250 device policies as well. So being able to put 00:21:12.250 --> 00:21:15.450 that into action, and I'm going to hamper on 00:21:15.450 --> 00:21:17.529 that phishing -resistant MFA, I think it's because 00:21:17.529 --> 00:21:20.470 we have that internally at Microsoft today, is 00:21:20.470 --> 00:21:22.549 really securing it through conditional access 00:21:22.549 --> 00:21:25.630 policies. Hey, this may sound like a really stupid 00:21:25.630 --> 00:21:28.730 question, but can we block someone completely 00:21:28.730 --> 00:21:31.930 if they're not using phishing -resistant MFA? 00:21:33.029 --> 00:21:35.630 Great question, Michael. Yes, you can. You can 00:21:35.630 --> 00:21:39.269 set a conditional access policy. So for example, 00:21:39.309 --> 00:21:42.250 you can target specifically to admins. It'll 00:21:42.250 --> 00:21:45.630 be like, hey, you need to use at least phishing 00:21:45.630 --> 00:21:48.309 -resistant MFA, like a FIDO2 key, right? So let's 00:21:48.309 --> 00:21:50.650 talk about FIDO2 key. If this person doesn't 00:21:50.650 --> 00:21:53.029 have it, you know what? you can't have access 00:21:53.029 --> 00:21:55.950 to this resource. But I do want to emphasize 00:21:55.950 --> 00:21:58.430 that you have to be careful when you put in these 00:21:58.430 --> 00:22:01.269 conditional access policies in because you can 00:22:01.269 --> 00:22:04.309 select all of your users to put in that group. 00:22:04.450 --> 00:22:06.809 But also thinking about the group, the set of 00:22:06.809 --> 00:22:08.849 groups that you may not want to include. For 00:22:08.849 --> 00:22:11.190 example, you could have organizational emergency 00:22:11.190 --> 00:22:14.450 access or break glass accounts as well. So with 00:22:14.450 --> 00:22:16.349 conditional access policy, it's really good to 00:22:16.349 --> 00:22:19.509 kind of list out who are your sort of high targeted 00:22:19.509 --> 00:22:22.950 users and then think about other accounts that 00:22:22.950 --> 00:22:25.450 particularly may not need that um especially 00:22:25.450 --> 00:22:27.650 if you're gonna they break glass accounts always 00:22:27.650 --> 00:22:29.829 tend to be the top ones yeah i really want to 00:22:29.829 --> 00:22:31.789 stress something that's just so important here 00:22:31.789 --> 00:22:34.369 and that is people say oh you know i'm just going 00:22:34.369 --> 00:22:36.890 to receive multi -factor authentication on my 00:22:36.890 --> 00:22:41.509 phone using an sms message that's broken it is 00:22:41.509 --> 00:22:43.309 completely broken if you're a high enough profile 00:22:43.309 --> 00:22:46.779 target someone can easily be bribed. Look, I'm 00:22:46.779 --> 00:22:48.700 making up an example scenario, but we've seen 00:22:48.700 --> 00:22:50.779 it. If you're a high enough value target where 00:22:50.779 --> 00:22:54.019 someone can actually clone your SIM card or have 00:22:54.019 --> 00:22:56.579 someone, I hate to say it, but perhaps someone 00:22:56.579 --> 00:22:58.720 who has insider knowledge, and I'll just leave 00:22:58.720 --> 00:23:01.319 it at that, and be able to clone your SIM. And 00:23:01.319 --> 00:23:03.079 at that point, someone else is responding to 00:23:03.079 --> 00:23:05.460 your SMS messages. It is actually a real attack. 00:23:05.579 --> 00:23:07.299 People think, oh, it's all James Bond stuff. 00:23:07.559 --> 00:23:10.380 It's not. Your nine -year -old child may not 00:23:10.380 --> 00:23:13.200 be a target, but an executive or someone with, 00:23:13.839 --> 00:23:16.799 major access to the environment may very well 00:23:16.799 --> 00:23:19.559 be a target. So the world just needs to move 00:23:19.559 --> 00:23:21.920 away from these non -phishing resistant credentials 00:23:21.920 --> 00:23:24.779 and this non -phishing resistant MFA. Certainly 00:23:24.779 --> 00:23:27.460 for your high value targets, most organizations 00:23:27.460 --> 00:23:29.900 will have them. So, you know, it's not James 00:23:29.900 --> 00:23:32.599 Bond, it's real. So you touched on integration 00:23:32.599 --> 00:23:36.160 with a seam like Sentinel, but you also mentioned 00:23:36.160 --> 00:23:38.119 co -pilot, security co -pilot. Do you want to 00:23:38.119 --> 00:23:41.740 sort of elaborate on what you mentioned? Oh, 00:23:41.859 --> 00:23:44.640 love it. I think for now this is one of my favourite 00:23:44.640 --> 00:23:48.859 subjects. So for those who aren't familiar, Microsoft 00:23:48.859 --> 00:23:52.740 Security Co -Pilot, we went GA with the product 00:23:52.740 --> 00:23:55.680 of 1st of April of last year. No joke. We are 00:23:55.680 --> 00:23:58.660 coming up to a year of the product being in GA. 00:23:59.059 --> 00:24:03.839 So Microsoft Security Co -Pilot is a Gen AI powered 00:24:03.839 --> 00:24:08.079 security solution. And it's really targeted to 00:24:08.079 --> 00:24:11.740 security and IT professionals to respond to cyber. 00:24:12.329 --> 00:24:16.029 threats, process signals, assess risk exposure, 00:24:16.390 --> 00:24:20.750 sort of machine speed and scale. So you can use 00:24:20.750 --> 00:24:24.410 it as proactively as you can to be able to really 00:24:24.410 --> 00:24:27.869 get a hold of some of the attacks that could 00:24:27.869 --> 00:24:30.950 potentially come in. And along with that, it's 00:24:30.950 --> 00:24:34.210 natural language. So Copilot there is really 00:24:34.210 --> 00:24:38.609 as an assistive experience to help SOC analysts. 00:24:38.809 --> 00:24:42.750 But I want to expand this because we're We have 00:24:42.750 --> 00:24:46.630 multiple integrations with our first party at 00:24:46.630 --> 00:24:49.910 Microsoft. That's obviously Entra being one of 00:24:49.910 --> 00:24:53.950 them, but Intune, Defender, we have Azure WAF. 00:24:54.349 --> 00:24:56.269 There's so many products I could lose count, 00:24:56.509 --> 00:24:59.210 but we've seen that integration grow in. And 00:24:59.210 --> 00:25:01.549 it really is, when you'll hear me, there's kind 00:25:01.549 --> 00:25:04.390 of two differentials. Like we have the immersive, 00:25:04.470 --> 00:25:07.670 which is the standalone experience. And then 00:25:07.670 --> 00:25:09.650 we have the embedded experience, which is really 00:25:09.650 --> 00:25:11.990 the experience that you can have within the portal 00:25:11.990 --> 00:25:14.329 so say for example today that would be in the 00:25:14.329 --> 00:25:16.529 entry portal as well do you want to give me an 00:25:16.529 --> 00:25:18.839 example of what a query might look like Okay, 00:25:19.019 --> 00:25:21.660 so say if I'm in, it could be either, right? 00:25:21.720 --> 00:25:24.480 So I really get to say could in either. Say if 00:25:24.480 --> 00:25:27.220 a SOC analyst wants to go in, or it could be 00:25:27.220 --> 00:25:29.000 an identity analyst, it could be an identity 00:25:29.000 --> 00:25:32.440 admin, and they want to know, like, who are the 00:25:32.440 --> 00:25:35.200 high -risk users? So you can go directly into 00:25:35.200 --> 00:25:37.559 the experience, type it in, literally as you 00:25:37.559 --> 00:25:40.440 would, and I would really recommend be as verbose 00:25:40.440 --> 00:25:43.140 as you want. Say it's a one -on -one conversation 00:25:43.140 --> 00:25:45.059 that I'm having with either Gladys or Michael. 00:25:45.259 --> 00:25:48.160 I want to be as descriptive as possible. So if 00:25:48.160 --> 00:25:51.059 I go in, pop in, say, I would like to know the 00:25:51.059 --> 00:25:54.640 top five risky users identified as high risk 00:25:54.640 --> 00:25:57.940 within my tenant, and I'm a big table person, 00:25:58.160 --> 00:26:00.180 I'll ask for it to be converted into a table. 00:26:00.460 --> 00:26:03.519 What security co -pilot will do, it will reason 00:26:03.519 --> 00:26:06.359 over, like, okay, what is Sharon asking me? Now 00:26:06.359 --> 00:26:08.640 I need to differentiate. Let me pull that, and 00:26:08.640 --> 00:26:10.839 then we're going to see which. where a plug -in 00:26:10.839 --> 00:26:13.640 that is okay that's the entra plug -in it will 00:26:13.640 --> 00:26:16.359 then go to the orchestrator figure out exactly 00:26:16.359 --> 00:26:18.460 what needs to do and i do want to emphasize that 00:26:18.460 --> 00:26:21.579 we do have responsible ai uh catering around 00:26:21.579 --> 00:26:24.000 that before you actually get a full response 00:26:24.000 --> 00:26:26.619 and then it will bring back a response to me 00:26:26.619 --> 00:26:30.359 and say these are all your five top users and 00:26:30.359 --> 00:26:33.460 hopefully there aren't five, there may be three. 00:26:33.539 --> 00:26:35.859 That's the big hope. But you'll get that list 00:26:35.859 --> 00:26:38.220 and you can basically share that information 00:26:38.220 --> 00:26:41.700 as well. So within Copilot today, within the 00:26:41.700 --> 00:26:43.900 actual standalone experience or the immersive 00:26:43.900 --> 00:26:47.599 experience, you have workspaces. So you can actually 00:26:47.599 --> 00:26:51.099 share that with another analyst if there is actually 00:26:51.099 --> 00:26:52.839 like an investigation that's going like, hey, 00:26:52.859 --> 00:26:54.839 look, I've actually found this. Let me share 00:26:54.839 --> 00:26:57.079 this with you. Or if you want to export that 00:26:57.079 --> 00:26:59.039 information. can also export that information 00:26:59.039 --> 00:27:02.000 as well. If there's some analysts that are working 00:27:02.000 --> 00:27:06.000 together and one analyst has access to some data, 00:27:06.039 --> 00:27:08.940 but the other doesn't, how do they share across? 00:27:09.519 --> 00:27:13.980 Can the second analyst have access to that information? 00:27:14.980 --> 00:27:17.900 So in Security Copilot today, we're actually 00:27:17.900 --> 00:27:22.140 using the RBAC model. So say if Analyst A is 00:27:22.140 --> 00:27:25.740 an identity admin and they have access to just 00:27:25.740 --> 00:27:27.960 the Entra portal and they don't have access, 00:27:28.240 --> 00:27:30.960 say, for example, to Intune and they need to 00:27:30.960 --> 00:27:34.000 share information, you can actually do that through 00:27:34.000 --> 00:27:37.339 workspaces. So if there's a workspace, you're 00:27:37.339 --> 00:27:39.859 like, okay, I need to give, I'd say, for example, 00:27:39.859 --> 00:27:42.759 I'm running an investigation as an admin. Michael 00:27:42.759 --> 00:27:44.799 goes, I need to share that information. with 00:27:44.799 --> 00:27:47.480 Michael. I can go through the workspace. I can 00:27:47.480 --> 00:27:49.559 put his access on top of that and say, hey, there 00:27:49.559 --> 00:27:51.579 you go, Michael. Here's the information that 00:27:51.579 --> 00:27:53.940 I found in this particular incident, which I 00:27:53.940 --> 00:27:56.640 feel would be pertinent for you when you're investigating 00:27:56.640 --> 00:27:59.440 which device that you're looking at as well. 00:27:59.559 --> 00:28:02.660 This was a huge ask. I just want to kind of emphasize 00:28:02.660 --> 00:28:05.680 a huge ask from our customers to be able to share 00:28:05.680 --> 00:28:08.819 workspaces and to also be complicit with the 00:28:08.819 --> 00:28:10.700 RBAC models that they're actually using within 00:28:10.700 --> 00:28:13.000 their organizations. And that's all audited, 00:28:13.000 --> 00:28:16.339 right? That is correct. Yes, 100 % is all. Yeah, 00:28:16.359 --> 00:28:18.039 that's incredibly important when it comes to 00:28:18.039 --> 00:28:21.440 sharing stuff. Cool. So Sharon, let's bring it 00:28:21.440 --> 00:28:25.940 all together. How would you put this with an 00:28:25.940 --> 00:28:28.539 example? What exactly are we trying to protect 00:28:28.539 --> 00:28:34.279 and what tools or capabilities we're using through 00:28:34.279 --> 00:28:37.539 that protection? I love that, Gladys. So when 00:28:37.539 --> 00:28:40.900 we think about scenarios, you know, there's so 00:28:40.900 --> 00:28:42.779 many out there right now. If we, you know, if 00:28:42.779 --> 00:28:45.180 you peep your head into... cybersecurity news 00:28:45.180 --> 00:28:47.599 that's out there. We've had, there's companies 00:28:47.599 --> 00:28:50.640 that are getting breaches. I think as those, 00:28:50.779 --> 00:28:53.539 we were just talking about this today, actually, 00:28:53.579 --> 00:28:56.740 in one of our team meetings, how, you know, there's 00:28:56.740 --> 00:28:59.039 health professionals that are particularly, you 00:28:59.039 --> 00:29:02.339 know, they'll be using chat GTP, sharing sensitive 00:29:02.339 --> 00:29:05.359 information. There could also be scenarios where 00:29:05.359 --> 00:29:08.119 people are using sort of the external tools, 00:29:08.339 --> 00:29:11.240 could be IP, sort of intellectual protection. 00:29:12.970 --> 00:29:14.930 the details that they're sharing in chat DTP 00:29:14.930 --> 00:29:18.269 as well. So when we kind of think about AI systems 00:29:18.269 --> 00:29:21.950 and how to protect it from an identity or even 00:29:21.950 --> 00:29:25.269 devices and looking at security frameworks is 00:29:25.269 --> 00:29:28.269 with monitoring, you're continuously tracking 00:29:28.269 --> 00:29:31.430 AI interactions, right? That's user access and 00:29:31.430 --> 00:29:34.910 system behavior to identify anomalies, preventing 00:29:34.910 --> 00:29:38.549 security breaches and enforcing policies. And 00:29:38.549 --> 00:29:40.970 the governance piece is really establishing security 00:29:40.970 --> 00:29:43.990 frameworks. So that's where we talked about role 00:29:43.990 --> 00:29:47.109 -based access controls, compliance rules, and 00:29:47.109 --> 00:29:50.910 data protection policies to ensure that AI operates 00:29:50.910 --> 00:29:55.829 securely. massive emphasis on ethical right that's 00:29:55.829 --> 00:29:59.349 always a like hey am i sure that the information 00:29:59.349 --> 00:30:01.710 that i'm actually getting from ai is accurate 00:30:01.710 --> 00:30:05.529 as well um so when we think about entra and what 00:30:05.529 --> 00:30:08.089 it does is to kind of put it all together enter 00:30:08.089 --> 00:30:11.190 provides that robust identity governance framework 00:30:11.190 --> 00:30:15.170 from least privileged access to automated account 00:30:15.170 --> 00:30:19.859 life cycle management doing granular role assignments. 00:30:20.099 --> 00:30:21.599 And if we think about Azure, that stands for 00:30:21.599 --> 00:30:24.440 management groups, subscriptions, and resource 00:30:24.440 --> 00:30:27.160 groups, and even looking at sort of resource 00:30:27.160 --> 00:30:31.039 levels. So that's really ensuring that precise 00:30:31.039 --> 00:30:35.079 access control. You can block apps. You can go 00:30:35.079 --> 00:30:36.660 in and say, actually, do you know what? I don't 00:30:36.660 --> 00:30:39.710 want... anybody at this organization to be accessing 00:30:39.710 --> 00:30:43.190 chat tdp to you know to be taking that information 00:30:43.190 --> 00:30:46.269 that's internal that's under nda we shouldn't 00:30:46.269 --> 00:30:49.269 be putting that into external tools as well um 00:30:49.269 --> 00:30:51.910 really with access reviews we've touched on it 00:30:51.910 --> 00:30:54.450 integrated that we have today is integrated with 00:30:54.450 --> 00:30:57.579 ml driven insight so it's continuously evaluating 00:30:57.579 --> 00:31:02.819 and optimizing permissions for AI applications. 00:31:02.859 --> 00:31:06.920 So reducing privilege creep and mitigating those 00:31:06.920 --> 00:31:10.940 security risks as well. So really, if we were 00:31:10.940 --> 00:31:14.420 to think about it, monitoring all of this and 00:31:14.420 --> 00:31:17.200 monitoring the capabilities of creating applications. 00:31:18.729 --> 00:31:21.029 organizations, you know, can achieve that sort 00:31:21.029 --> 00:31:24.309 of zero trust element of identity management. 00:31:24.410 --> 00:31:28.630 So reducing the attack surface and the AI workloads 00:31:28.630 --> 00:31:31.869 protection that you can have with Entra ID. Yeah, 00:31:31.910 --> 00:31:33.490 there's a couple things I want to add here. First 00:31:33.490 --> 00:31:34.869 of all, I actually know quite a few companies 00:31:34.869 --> 00:31:37.349 who have actually blocked access to chat GPT 00:31:37.349 --> 00:31:39.950 and other tools of that ilk as well because they're 00:31:39.950 --> 00:31:42.589 public and you don't want to start, you know. 00:31:43.279 --> 00:31:45.099 start, hey, we're about to acquire a company, 00:31:45.240 --> 00:31:47.140 please give me some, you know, some wording to 00:31:47.140 --> 00:31:49.099 use in the cover letter, right? And now all of 00:31:49.099 --> 00:31:51.579 a sudden you've disclosed potentially to somebody 00:31:51.579 --> 00:31:53.180 that, you know, you're about to purchase a company. 00:31:53.640 --> 00:31:55.660 So, yeah, so I know a lot of companies that have 00:31:55.660 --> 00:31:58.319 actually blocked access to, you know, large language 00:31:58.319 --> 00:32:01.940 model tools like ChatGPT and so on, and rather 00:32:01.940 --> 00:32:04.960 use tools like, you know, Azure AI, you know, 00:32:04.960 --> 00:32:08.440 using... their own subscription. That way the 00:32:08.440 --> 00:32:10.359 data stays within their own environment and is 00:32:10.359 --> 00:32:13.119 not used for any other purpose. The other thing 00:32:13.119 --> 00:32:15.099 I really want to point out is there's multiple 00:32:15.099 --> 00:32:19.119 ways of... mitigating you know these sort of 00:32:19.119 --> 00:32:21.900 defense these sort of issues and a really interesting 00:32:21.900 --> 00:32:25.640 way of kind of looking at it is you kind of have 00:32:25.640 --> 00:32:27.940 preventative controls and then detective controls 00:32:27.940 --> 00:32:30.319 and responsive controls and so for the detective 00:32:30.319 --> 00:32:32.339 control so preventative controls which your first 00:32:32.339 --> 00:32:33.960 one is just like our back right just blocking 00:32:33.960 --> 00:32:36.589 someone from doing something preventing access 00:32:36.589 --> 00:32:39.369 to chat GPT and other tools, right? So they are 00:32:39.369 --> 00:32:42.009 all preventative controls. And least privilege 00:32:42.009 --> 00:32:44.089 is another example of that. And then you have 00:32:44.089 --> 00:32:46.250 detective controls, which is your monitoring, 00:32:46.490 --> 00:32:48.750 your logging, your auditing, your copilot going 00:32:48.750 --> 00:32:50.869 into query for certain things, all that sort 00:32:50.869 --> 00:32:52.849 of stuff. And then finally, you have responsive 00:32:52.849 --> 00:32:55.950 controls where you respond to something happening. 00:32:56.450 --> 00:32:59.109 So it's always, I think, very useful to think 00:32:59.109 --> 00:33:01.289 about what sort of controls are at your disposal 00:33:01.289 --> 00:33:04.029 in those sort of two or three, I should say, 00:33:04.029 --> 00:33:06.609 major buckets. I want to add something else, 00:33:06.609 --> 00:33:10.269 Michael, to what you just said. Even governance, 00:33:10.509 --> 00:33:14.809 right? Data governance. You have all the sensitive 00:33:14.809 --> 00:33:19.569 labels, whether it's confidential or public data, 00:33:19.730 --> 00:33:24.890 which adds encryption to the data itself. Your 00:33:24.890 --> 00:33:28.009 DLP type of policies. And we're going to talk 00:33:28.009 --> 00:33:30.950 more about these during the governance session. 00:33:31.500 --> 00:33:35.500 But this is important because now an attacker 00:33:35.500 --> 00:33:40.059 can basically, if they get an identity, they 00:33:40.059 --> 00:33:43.220 could impersonate this person and just query 00:33:43.220 --> 00:33:47.859 to AI. So now you are asking information to the 00:33:47.859 --> 00:33:50.640 AI. This attacker may be asking information to 00:33:50.640 --> 00:33:53.920 the AI and acquiring that information without 00:33:53.920 --> 00:33:58.019 actually having to do things like reconnaissance 00:33:58.019 --> 00:34:02.089 in the environment. So it is important, all this 00:34:02.089 --> 00:34:04.509 monitoring to be in place, and it's important 00:34:04.509 --> 00:34:08.570 for all the data protections. And we talk about 00:34:08.570 --> 00:34:11.570 our bioconditional access to be in place. Yeah, 00:34:11.590 --> 00:34:14.389 exactly. They do different things. They're all 00:34:14.389 --> 00:34:16.670 incredibly important. But you do need to sort 00:34:16.670 --> 00:34:20.130 of have mitigations or technology in each of 00:34:20.130 --> 00:34:21.670 those buckets as well, right? You can't just 00:34:21.670 --> 00:34:24.889 rely on monitoring. Well, you can, but you better 00:34:24.889 --> 00:34:27.690 hope that, you know. you know you're catching 00:34:27.690 --> 00:34:29.750 the right things but yeah you got to have the 00:34:29.750 --> 00:34:32.170 you know the sort of the preventative controls 00:34:32.170 --> 00:34:34.909 as well as the detective controls as well do 00:34:34.909 --> 00:34:36.409 you want to add anything else there sharon or 00:34:36.409 --> 00:34:38.889 are we just are we done with that topic no i 00:34:38.889 --> 00:34:41.349 think um i think we're done i think we're done 00:34:41.349 --> 00:34:43.170 with that actually do you know no there is there 00:34:43.170 --> 00:34:45.579 is one thing right so i think what you've added 00:34:45.579 --> 00:34:49.340 on in terms of everything, all the actions. And 00:34:49.340 --> 00:34:52.119 I love the emphasis on proactive, right? So we 00:34:52.119 --> 00:34:57.059 find in many cases where it might be too late. 00:34:57.159 --> 00:34:59.480 You already know an organization's already been 00:34:59.480 --> 00:35:02.420 sort of breached. There could be an attack. The 00:35:02.420 --> 00:35:04.739 information could already be shared with some 00:35:04.739 --> 00:35:07.980 of these shadow sort of IT elements of using 00:35:07.980 --> 00:35:12.119 external tools. So I really want to emphasize 00:35:12.119 --> 00:35:15.820 on that. just making sure that you have the right 00:35:15.820 --> 00:35:19.639 insights. And along with that, with these tool 00:35:19.639 --> 00:35:22.960 sets, being able to reduce that manual investigation 00:35:22.960 --> 00:35:26.320 effort. So the key thing here is that speeding 00:35:26.320 --> 00:35:29.699 up that response, you know, and getting real 00:35:29.699 --> 00:35:32.119 -time efforts as well, especially during around 00:35:32.119 --> 00:35:34.539 application threats that the AI can introduce 00:35:34.539 --> 00:35:39.099 within an environment. So Sharon, how... does 00:35:39.099 --> 00:35:43.019 a typical day in Sharon's life look like? And 00:35:43.019 --> 00:35:48.030 I mean, in Microsoft work life. Ah, Gladys, it's 00:35:48.030 --> 00:35:50.929 hectic. If I was to use one word, it's hectic. 00:35:51.630 --> 00:35:54.429 But at the same time, it's also very exciting. 00:35:54.889 --> 00:35:57.969 The one thing I could guarantee is that I am 00:35:57.969 --> 00:36:01.050 learning something new every day. I really enjoy 00:36:01.050 --> 00:36:04.190 what I do in terms of the team that I'm in. I 00:36:04.190 --> 00:36:06.909 interact with organizations every day. I'm having 00:36:06.909 --> 00:36:10.349 a conversation, especially around, obviously, 00:36:10.469 --> 00:36:13.110 the security co -pilot pieces and along with 00:36:13.110 --> 00:36:16.079 Entra. And also, you know, I get to work and 00:36:16.079 --> 00:36:17.920 collaborate very closely with our engineering 00:36:17.920 --> 00:36:21.460 teams. It's super important for us to listen 00:36:21.460 --> 00:36:24.639 to our customers, especially when we're trying 00:36:24.639 --> 00:36:28.159 to support their adoption within their environments 00:36:28.159 --> 00:36:31.340 as well. You know, while I do that, I also have 00:36:31.340 --> 00:36:35.119 the pleasure of managing eight people as well. 00:36:35.199 --> 00:36:38.340 Wonderful people, might I add. So it definitely 00:36:38.340 --> 00:36:43.360 adds a variety to my day to be able to help them 00:36:43.360 --> 00:36:46.489 and coach them and guide them so that's definitely 00:36:46.489 --> 00:36:49.809 an element that i enjoy doing as part of my role 00:36:49.809 --> 00:36:52.730 you know it's interesting you know we've had 00:36:52.730 --> 00:36:54.949 people on the on the podcast before we've asked 00:36:54.949 --> 00:36:56.510 them the same question about you know what does 00:36:56.510 --> 00:36:58.449 a typical day look like and even though we ask 00:36:58.449 --> 00:36:59.769 people what does a typical day look like it's 00:36:59.769 --> 00:37:01.590 funny how many people say they just love what 00:37:01.590 --> 00:37:04.110 they do i think that's something i'm sure a lot 00:37:04.110 --> 00:37:05.849 of companies are the same but you know i've been 00:37:05.849 --> 00:37:09.360 here now for nearly 33 years and You know, you 00:37:09.360 --> 00:37:11.199 get to work with some really smart people. You 00:37:11.199 --> 00:37:13.559 think you're the smartest person in the room 00:37:13.559 --> 00:37:16.420 and 99 times out of 100, you're not. There's 00:37:16.420 --> 00:37:18.219 a lot of really clever people you can talk to 00:37:18.219 --> 00:37:22.480 who are very insightful, you know, not just smart, 00:37:22.579 --> 00:37:25.019 but also making a huge difference. And yeah, 00:37:25.079 --> 00:37:26.519 it's great to be able to work with people like 00:37:26.519 --> 00:37:30.000 that on a daily basis. All right. So one thing 00:37:30.000 --> 00:37:34.019 we always ask our guests is if you had one final 00:37:34.019 --> 00:37:35.860 thought to leave our listeners with, what would 00:37:35.860 --> 00:37:39.400 it be? So I think in terms of, you know, when 00:37:39.400 --> 00:37:43.360 we think of Gen AI, it's rapid, right? There's 00:37:43.360 --> 00:37:46.000 always a new module that's coming out. I open 00:37:46.000 --> 00:37:48.659 up my news app or I can open up LinkedIn and 00:37:48.659 --> 00:37:50.940 it's like, well, OK, I'm oversaturated with so 00:37:50.940 --> 00:37:54.539 much information. And it's a stark reminder that 00:37:54.539 --> 00:37:57.500 when we think about how we can protect ourselves, 00:37:57.800 --> 00:38:00.980 our organizations and even our families. Right. 00:38:01.039 --> 00:38:02.800 So because people could download apps on their 00:38:02.800 --> 00:38:08.349 phone is. We think about how it's crucial to 00:38:08.349 --> 00:38:13.269 make sure that AI is operating safely. Real focus 00:38:13.269 --> 00:38:17.329 on ethical as well. And also, you know, when 00:38:17.329 --> 00:38:19.130 we talk about organization, there's alignment 00:38:19.130 --> 00:38:21.489 with organization and even regulatory standards 00:38:21.489 --> 00:38:25.349 as well. Continuous oversight, transparency, 00:38:25.769 --> 00:38:29.869 adaptability are really key to mitigate risks 00:38:29.869 --> 00:38:33.909 like bias, security threats, and even thinking 00:38:33.909 --> 00:38:37.119 about model drifts as well. So if we think about 00:38:37.119 --> 00:38:39.659 integrating the real -time monitoring, think 00:38:39.659 --> 00:38:41.940 about your access controls, think about clear 00:38:41.940 --> 00:38:45.980 accountability structures and how you can train 00:38:45.980 --> 00:38:49.539 or how you can maintain trust and reliability 00:38:49.539 --> 00:38:52.900 in AI systems as well. And, you know, one thought 00:38:52.900 --> 00:38:55.599 is that how do we keep up with the trends of 00:38:55.599 --> 00:38:58.800 emerging threats? Listen to podcasts. The Azure 00:38:58.800 --> 00:39:02.559 podcast is one. There's so many out there. It 00:39:02.559 --> 00:39:05.559 can be a struggle to keep up to date, but I think 00:39:05.559 --> 00:39:07.179 that's where you can use the power of Gen AI. 00:39:07.619 --> 00:39:10.480 Go in and say, hey, give me the top 10 news articles 00:39:10.480 --> 00:39:12.260 that I need to be looking at when it comes to 00:39:12.260 --> 00:39:14.619 threats to Gen AI and be as descriptive as possible. 00:39:15.539 --> 00:39:17.480 Yeah, you've got to stay on top of things, right? 00:39:17.619 --> 00:39:20.440 I mean, again, as I mentioned before, I've been 00:39:20.440 --> 00:39:22.260 at Microsoft for a long time, yet I am still 00:39:22.260 --> 00:39:24.039 learning every single day and actually make it 00:39:24.039 --> 00:39:26.349 a priority of mine as well. But that's a discussion 00:39:26.349 --> 00:39:28.449 for another day. All right, well, Sharon, thank 00:39:28.449 --> 00:39:30.530 you so much for joining us this week. I know 00:39:30.530 --> 00:39:33.909 you're very busy, so I always appreciate anyone 00:39:33.909 --> 00:39:37.590 taking the time to come on the podcast. And as 00:39:37.590 --> 00:39:39.550 usual, I always learn something. I know Gladys 00:39:39.550 --> 00:39:40.630 probably learned a thing or two, even though 00:39:40.630 --> 00:39:42.030 she knows a lot more about this stuff than I 00:39:42.030 --> 00:39:44.289 do. And to all our listeners out there, we hope 00:39:44.289 --> 00:39:46.849 you found this episode useful. Stay safe, and 00:39:46.849 --> 00:39:51.579 we'll see you next time. You can find show notes 00:39:51.579 --> 00:39:55.579 and other resources at our website, azsecuritypodcast 00:39:55.579 --> 00:39:59.340 .net. If you have any questions, please find 00:39:59.340 --> 00:40:02.780 us on Twitter at AzureSecPod. Background music 00:40:02.780 --> 00:40:06.119 is from ccmixter .com and licensed under the 00:40:06.119 --> 00:40:07.179 Creative Commons license.