Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Microsoft Cloud IT Pro Podcast

Microsoft Cloud IT Pro Podcast artwork

Episode 434: Reading the Spec Sheet on Microsoft's Agentic SOC

Aug 13, 202648 minAudio

Welcome to Episode 434 of the Microsoft Cloud IT Pro Podcast. Microsoft announced Project Perception on 2026-07-27 with a lot of “workforce of AI agents” language. In this episode, Ben and Scott discuss what Perception really is, what it needs from your tenant, what it costs, and what happens the first time an agent asks you to approve something. Six agents. Seven playbooks. Every one manually triggered. Each agent gets its own Entra Agent ID and a permission grant. There’s a lot to unpack. Show Notes Save $200 at TechCon365 with code Ben200: https://www.techcon365.com/Seattle/ What is Project Perception? How Nationwide stays ahead of attackers with Project Perception Key concepts in Project Perception Agent categories in Project Perception Microsoft Security Copilot Security Compute Units and capacity Work with playbooks in Project Perception Prior podcast context Ep 431: Agent Governance Is the New App Governance Ep 419: Security and AI Ep 416 and Ep 412 Ep 406: Agents of Insight Ep 377: Microsoft Copilot for Security About the sponsors Nasuni is a leading unstructured data platform for enterprises where file data is mission-critical for both people and AI. Nasuni powers the operational file layer where work happens - helping organizations manage, protect, and activate data so teams can work smarter, reduce costs, and operate securely without limits. Intelligink - Would you like to become the irreplaceable Microsoft 365 resource for your organization? Let us know!

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

- Welcome to episode 434 - of the Microsoft Cloud IT Pro podcast recorded - live on 08/10/2026. - This is a show about Microsoft three sixty - five and Azure from the perspective of IT - pros and end users, where we discuss a - topic or recent news and how it relates - to you. In this episode, Ben and Scott - provide their perspective on Project Perception, - Microsoft's latest AI security tool that provides you - with a workforce of AI agents. We discuss - what perception really is, what you need to

- have to get started, and the agents and - tools that are contained within the product. - Join us as we unpack - Project Perception. - Scott, I was just I mentioned this to - you. I had a statement - that I made two weeks ago when we - recorded on the whole authentication and passkeys, and - we mentioned that you can go discuss with - us on Reddit. And somebody was like, hey, - Ben. You said something incorrect, and they called - me out on Reddit. So, yes, I made - a couple of comments. One, I said I

- thought synced pass keys were tied to authenticator, - and I had it backwards in my head. - The device bound our authenticator. You can't do - synced pass keys in authenticator. - So, - yeah, I had it backwards in my head, - and - we were talking about one password. I also - discovered one password does work now for synced - pass keys. I just had a misconfiguration - of my tenant that was preventing me from - doing it. I think I have misconfiguration of - my tenant or my employees. I can you - make configuration - changes in your tenant? Not like that. They - don't let you go just add random

- passkey - GUIDs to I don't even know. Is it - GUIDs? Is it GUIDs? Is it yeah. I - say GUID. You say GUID. I interchange them - sometimes. - You've got that whole Pan Panhandle thing going - on. Like, there there's a there there's a - little bit of an accent from the Panhandle. - Yeah. Usually, I say GUID, and then every - once in a while, I get that GUID - that sneaks in there. No. The AA GUID, - they don't let you just add those randomly - to your tenant? Probably not. Do you have - agents - scanning your tenant to look for those types - of things? I'm sure. I do. But, I - mean, what is an agent?

- What is an agent? - What is a skill? What is AI? - Yes. You can pontificate. - Agent Smith? Agent - Smith? Oh, oh, no. We're gonna go Skynet. - Agent Smith. We're gonna go Skynet first. Oh, - Agent Smith. No. We can do Agent Smith. - Do you realize how old that movie is? - We talked about this before. Came out '99, - Scott. That that that We're going on twenty - seven years. Mhmm. - Yeah. Yeah. My my - we were watching a TV show the other - day, and - there was a line in there. It was

- a it was a teenager talking to we're - watching a movie, but it was a teenager - talking to her parent. - And - her dad was born in 1980, like, from - the nineteen hundreds Like kinda kinda Yeah. And - and and she asked this quipped to her - dad that - you were born closer to 1940 - than I was to 1980. - Meaning, like, when we were born, we were - closer to the forties than a kid today, - forty five, forty six years later, - would actually be to 1980. - That that was a really weird one for

- me. It's like I'm I'm old in multiple - dimensions. - Yeah. I don't know that I wanna think - about that too hard. Yeah. I just wanna - pretend I'm still 20. Yeah. Well, I I - don't know that I know this could tear - all all my gray hair and everything, but - I got one more for you. Listen. You - hear that? The sound of silence? I don't - hear the sound of the kids. Are back - in school. - I get my house back. I get my - office back. How do you get how does - it feel to be back in silence? - It would be okay, except I got this - puppy behind me. So if he starts screeching, - you you you'll know why. Either he didn't

- like the content or he really needed to - go outside. It is a cute new puppy. - Yeah. Alright. I will let you enjoy the - puppies. I'm not gonna enjoy the puppies. I - even had a good segue into agents. We - got distracted by puppies in square islands. Sorry. - I took you there. We'll get back. It's - okay. Human in the loop. Human in the - loop. We're back. Human striked. Okay. That was - quite the perception there, Scott, of - the - wow. Okay. So agents. Human in the loop, - a new product from Microsoft. - A new service? I don't know. At what - point in time, speaking of agents and skills,

- what's a product now and what's a service? - I don't know anymore. Who knows? Up is - down. Up is down. Left is right. But, - yes. Prod Project Perception? Project Perception. - What are we perceiving? Project Perception. What are - we perceiving? We are perceiving - tenant security. - Like, - this is an interesting one. It is a - new security - product or service from Microsoft - and right from - the website for Project Perception, a workforce.

- So now you aren't maybe this isn't even - a service or a product. It is a - workforce. - It's a workforce of specialized - AI agents, - red, blue, and green, - that reason across your security data tools, workflows - to expose gaps, investigate threats, remediate them, use - up the strategy, agents carry the load. So - this is an agentic - security - based service - from Microsoft - that you can - go in and enable at some point in - time. Right now, it's like a limited public

- preview, - so - very selective in who can get access to - it. But all the documentation, the websites, everything - else is out there so that you can - now - enable agents to secure your environment. - Can you though? - Ish. - What is an agent, Scott? - I think that's one of the biggest things - that I I ran into this. But before - we get, like, too deep into that, maybe - we should talk a little bit about - what Project Perception, like, proposes to be and

- what the agents it proposes to be - that are in this limited public preview that - are out there. And then we can maybe - come back a little bit to what is - an agent and - what does, like, supervision - look like - in particularly context of project perception and what - they've released in this first iteration here? Yeah. - And even some of the onboarding of this, - I had to pull together. - Again, limited public preview, - all over the documentation, things are liable to - change,

- But I had to pull together a few - different things - to figure out, like, what do you even - need or what does this look like from - an onboarding perspective? Because - you have the home page, and then when - you go down in the documentation and start - looking at it, it has - certain requirements. - So if this is something you're interested in, - just getting started with it and looking through - some of the prerequisites - is this is something that's gonna live in - the Microsoft Defender portal, so you'll need to

- have access to Defender, Microsoft Defender XDR Got - security center. Defender. Yes. I can keep coming - up with acronyms. Yeah. Security.microsoft.com. - You do need to have a unified - role based access - control, RBAC? RBAC. This is this was a - new acronym that was popping up all over - the documentation for this one. - Yes. That needs to be enabled across all - Defender workflow or workloads. - You have to be a security admin, security - reader. - And then even as you get into some

- of the agents then, it talks about, - I believe, with some of these agents, like - different licensing, you need to have - Defender for Cloud enabled to look at your - Azure - stuff. You need to have - certain prerequisites - for various workloads, so Defender for Office three - sixty five - plan two. If you wanna look at email - and collaboration - stuff, you need to have Defender for Cloud, - Defender for Containers. - If you're wanting to look at your Azure - cloud, - Defender or Entra ID p two, Defender for

- Identity, Defender for Cloud apps to look at - identity alerts. - So this is one part that for me, - it's like all through the documentation, it take - talks about the licensing and Urbach and all - of that. But then when you go back - to the actual product page and you scroll - all the way down and you look at - how is project perception priced, then it comes - up with you pay for only what you - use measured - in security computing units. - The mythical the magical - SCU. Yeah. So good old security Copilot - SCUs.

- So if this is something you're interested in, - much like some of the other things we've - talked about, this is not I would not - call this a low cost barrier to entry. - You're probably looking at you already have - Microsoft three sixty five e five - or at least a lot of the security - products from e five, and you're going to - be willing to pay for security compute units - to be able to go in and run - these agents. So that is my here's where - you go, how you wanna think about if - this is something you're interested in just from

- a pure pricing, licensing, - onboarding perspective in that respect. It's got a - lot of - moving pieces and parts, and I think you - can see probably where it's, like, limited public - preview because it's for those folks out there - who are a little bit more in, like, - token maxi land or I think just kinda - some of the bleeding edge, like, really interested - in the - frontier programs and other things, like, that are - available today. Yeah. So once you turn all - that on, once you have everything licensed and - enabled and permission set, then you can go - start looking at these agents. Well, I mean,

- I just get an AI that does it - all for me. Right? Yeah. You just tell - it go do your thing. Go be my - security team, my security workforce. Go be my - security workforce - indeed. - Well, - so so, yeah, you you get all this - stuff lit up. There is a lot of - licensing. - There is a lot of configuration. - I didn't see, like, configured this for me - agent as part of onboarding, so you do - have to go and kinda wrangle that, well, - altogether. But once you do get that going,

- you get these - categories of agents, like, I guess, broadly. Right? - They group them into Yep. Red team, blue - team, and - green team. So red team, - traditional - kind of penetration side of things, like, out - there to - identify - what's exploitable - and potentially - exploit it, blue team to - monitor for those threats. - When they see them come through. Like, can - they reconstruct them, get through the path, - close any gaps? Like, if there was something

- exploitable, like, hey. Let me come and close - those up real quick. And then green team - to - remediate - and harden - configuration. So - while blue team might be doing, like, active - monitoring, triaging alerts, things like that, green team's - out there kinda - constantly - analyzing your posture, - making sure to give you, like, a set - of - prioritized recommendations, - the things like that that are gonna help - you ultimately, like, harden that posture. So I

- think that's, like, all good on the surface. - And you go like, Well, that would be - kind of interesting. Right? Like, so I've got - this thing that can get out there, and - it can try and Yep. Poke holes in - my configuration - and see what can actually get through. I've - got another piece over here that can kinda - monitor things and potentially - operationalize - that monitoring or that observability - and go close those gaps real quick. And - then I've got something else on the back - end that's kinda sitting more in a constant - loop and basically doing continuous improvement and run,

- run, run, kinda like CICD - for your tenant configuration is what it sounds - like. I think it's very easy to say, - like, oh, hey. That's what it all reads - as. But, ultimately, - back to that original question of what is - an agent, - that that very much, like, - bleeds back into - what project perception - is here because, - like, that that that was the thing that - struck me. So you go out, you read - about it, you read the blog post, things - like that, you go through the docs, and - you're like, cool. And then you get down

- to it, and - the the things that I was saying of, - hey. Let's do the the whole red, blue, - green, and it's a little bit more automated. - Turns out it's really heavy on human in - the loop. Everything is actually manually triggered, - and and and effectively read only, like, with, - like, one or two caveats, but pretty much - across the board, read only. - And if that's the case - and you're kinda like, you're doing all the - supervision and you're doing all the monitoring and - it is very heavy on human loop, is

- it truly agentic, or is it more like - a set of capabilities that you've been provided - maybe as, like, a security engineer within your - organization - to go and kinda help you have additive - tools to understand your environment configuration, things like - that. I can see where it goes in - the future, but it's it's hard to to - today to to kinda - be able to, like, really wrangle it together. - And then I wonder back to the whole - pricing aspect of it. Like, is it worth - it? Like, what are you getting here? - I kinda walked away at the end. And - I know this is a little flippant of

- me, but, like, you walk away at the - end of, like, there's a bunch of, like, - scripts and things out there that help you - evaluate your environment already. Like, what's additive here - that you're getting out of it? I have - done some of this type of stuff even - using existing - AI. So I think seeing some of this - and having it built in was a little - intriguing to me. I think as I look - at this, - certain agents are probably - I probably get different value from different agents - in this. And maybe we dive into them - like the red team agent. Like you said,

- that's gonna be identifying the weakness. - This one to me is - I think there's some value here, - but I think you also have some limitations - here. This one, when you go look through - it, Microsoft has the recon agent that's on - the red team. There's multiple agents. There there's - Right. There's multiple agents under these different teams. - Yeah. Under the teams categories. Yes. So yeah. - So so as we go through this, it's - like, yeah. Red team has recon agent, blue - team has - four agents,

- and green team has one agent under it. - So you could also potentially see, like, additional - capabilities being introduced over time - and having that stuff come through. But yeah. - So so maybe we start with the recon - agent out there - scouting around. These are always interesting. Right? Recon - agent to me because it's simulating attack behavior, - trying to expose vulnerabilities - that - adversaries could exploit. It's looking at things like - mapping attack paths,

- looking at choke points in your configuration. It - does have vulnerability - code scanning in there. I think maybe that - one jumps out at me a little bit, - exposes detection gaps, but it's all simulated. - Right? And this is the part where you - start thinking of you can only get so - far, I think, in a simulation. You can - only hit an API so many times before - you actually would have to make a configuration - change, like, to go through the next step. - So while I think it can find - it's almost like it could find surface level

- stuff, but it has a limit to the - depth that it could go to. Because if - you have to expose one vulnerability, - how can it know that if it maybe - gets past the first layer, that it can - move laterally? - Or if it exploits the certain code vulnerability, - how does it know if it can keep - taking the next step? Like, this is maybe - that first line. Maybe some of the AI - stuff can kinda say, well, if I can - get past here, then I can go do - this, then this. It's - but I also think there's some value here.

- Some of the other things I've done with - some of my other AI tools and looking - at this is taking some of the correlation - too between like what's known out there going - out looking at the web. What are known - CVEs out there? What are known exploits? What - are some of those things that are known - to be exploitable? - Do you see them in the configuration of - my tenant where an exploit from over here - could be used over here in my tenant? - The red team to me, a lot of - the investigatory - probing, - I think there's a level, but I also

- think there's a lot of - correlations and maybe looking at things from different - angles - where you can get some value from AI - here. I think it's back to the whole - what's an agent thing. I think it's interesting - that they called it an agent because, like - like, there is I'm not arguing that there's - not value in it, but I think the - presentation's interesting. - So if you go and read the docs - and you start hopping in and you say, - okay. What's the recon agent and what's it - do? It's very clearly positioned - as - a single pane of glass

- across - multiple Defender products. Right? So so this is - additive to Defender. So Defender is kinda the - base. It's out there. It's what protects you. - So so that's number one is it's a - little kinda like it is that little canned - reporting kinda thing, right, of saying, hey. Let's - go down the path, analyze your environment configuration. - I think where the AI thing comes in - and where it's a little bit less of - a canned report, but it's still very much - a report at the end of the day, - is the ability to go look at your - specific tenant configuration. - Let's say, the configuration - for

- maybe one of your subscriptions in Azure. - It's able to hop through that subscription specifically - and kinda do - more - more pivots - and kinda what if analysis - than a traditional, - alright, go check this binary, if this, then - that, - kind of thing. I think that's where AI - helps out these things. It makes it a - little bit more of a choose your own - adventure. - But, ultimately, getting to - that single - pane of glass - that that's out there, I think even that - single pane of glass is

- interesting the way it's kinda presented - back to you. So, like, when you go - and run these things, - you you don't get a traditional dashboard. Like, - it's not like you get, like, a Power - BI report or anything like that. You get - a bunch of markdown files in a kinda - known format. But if you were to run - it against my tenant and then run it - against your tenant, we should get different sets - of we get the same set of output - mark markdown files, but with different guidance in - each one. The markdown files to me makes - sense because - one of the other things, like, if we

- wanna go take this the next step, is - you get all this markdown. - It does talk about, and this is maybe - where some of that value comes in with - project - perception, is it can take the output of - one agent and feed it to the next - agent. Which you kinda need. Right? Like, if - you're gonna go and say, hey. I found - something wrong in your configuration. You kinda wanna - pass that off to the next thing to - have it evaluate. Say, like, okay. Is this - important? Where does it land? And that's where - kinda, like, the blue and green come in. - We'll just say, hey. Let's go what's prioritization

- look like here? Like, that's the responsibility of - the green team agent. It's not the responsibility - of any of the blue team agents that - are out there that are kinda doing active - triage or things like that. Yeah. And that's - where that red team it's like, we've gone - in. We've done the recon. We figured out - all these things. Now let's pass it over - to that blue team agent. - And this one, there's the most in here - where you have - four agents in the blue team. You have - the triage agent, you have the attack - investigation - agent, - the threat intelligence agent, - and the detection

- authoring agent. And I think this is one - where that red team agent can then take - those marks down to the blue team and - pass it over to the triage agent, where - this agent can now - go - take all of that - information. - This is where - I have mixed feelings on this because - the triage agent it would be interesting to - see how it passes. The triage agent has - defined classifies and triages as alerts in Defender - for different workloads, for different alert types. - It does have some things like autonomous

- triage. It has some learning based on feedback, - record classification. - But if your - red team agent - is read only, - it shouldn't, in theory, actually trigger any alerts - in Defender because it's not taking any actions, - I don't think. Maybe it would. Maybe it - would try - like, what is a read only agent, I - guess, Scott? Is it actually gonna try to - log in? Is it gonna try, like, brute - forcing a password a little bit? In my - mind, that's not really read only. It's doing

- a little bit more writing. - And to really get things to trigger in - Defender, - you have to do that type of stuff. - It would be nice to see this evolve - a little bit where - maybe the red the recon agent is allowed - to poke at your tenant a little bit - more. Microsoft's running it all, it would be - able to recognize it and flag it accordingly - so that blue team agent could actually - like, you could see a little bit more - of a simulated attack versus just a, let's - go read all the configuration. It's what I

- was kind of planning for. It was like, - hey. When do these things get out? What - if we get a little more proactive? But - e even touching things can be a little - touchy sometimes when they were not there. So - if you look at these agents and so - so per agent, there's a configuration, - and that configuration comes with a set of - permissions - that are gonna be layered into the graph - and your tenant and other things that are - out there. So - if you look at, like, the recon agent, - it's a - it's it gets some pretty privileged

- role assignments. Like, it does get user read - all. It gets group member read all, application - read all, delegated permissions read all, audit log - read all, - things that are out there. It also gets - a bunch of Azure RBAC assignments. - So you have to kinda go through, add - a subscription scope, and - assign these things down or assign it at - a higher level, like a management group, something - like that. Let it bleed through. But it - gets reader, - specifically Key Vault reader and Log Analytics reader, - as well as - security reader that are out there. So I

- think even going through and just kinda poking - at things, maybe it'll set some things off - depending on how your environment's configured and what - it is. But Right. I really look at - like, today, like, the red team - agent would be out there analyzing your environment. - It's gonna, like, tell you - what the broad scope of maybe think about, - like, a funnel. Like, you're always gonna be - funneling it down. So broad set of things - that are out here. The blue team is - gonna be sitting out there and saying, like, - okay. What are my tasks I need to - go? What do I need to triage that's

- out there? If I and this is where - maybe, like, even, like, the red agent goes - away a little bit is, like, you have - things like an attack investigation agent. So that - could be super helpful, right, because you've got - an active tenant. And you're out there. You - don't need the red team to come in - Yep. And look at things like that, tell - you about misconfigured - users, - anonymous logins, - all all those kinds of things. - So, yeah, you get that funnel, pass it - to blue team, and then, you get a - little bit more of the the the green - team expertise to say, like, hey. Here's what - we're gonna prioritize - to go and fix

- that's out there, which is interesting. Like, the - green team is really focused on prioritization. Like, - there's one agent today. It's the posture prioritization - agent. - Like, - like, the says it on the tin where - it's gonna go and what it's gonna do - for you. - If you are managing file storage today, you've - probably felt the pressure. Refresh cycles are harder - to plan, costs and complexity are increasing, especially - as file workloads scale across regions. - That's why Microsoft customers trust Nasuni as an

- Azure certified software. - Nasuni provides a global file system designed for - enterprise scale environments, - keeping file data secure, governed, and accessible across - Azure environments without added infrastructure complexity. - It's the foundation to keep your unstructured data - stable even as your infrastructure evolves. To see - how much you could save while keeping your - initiatives on track, visit nasuni.com/tco. - That's nasuni.com/tco.

- I like the attack investigation agent. That if - I was gonna just from the title without - having played with these, look at these, I - can see a lot of value with the - attack investigation - agent. - I would like it to see it get - to a better price point. You need a - lot for this one. Right? Can you come - back to what do you need to turn - on to get all this stuff going? This - one's like XDR, - EU fives - for all the categories of attacks, identity attacks, - and - kind of cloud and compute side of things.

- Kinda looking at a lot. Right? The cloud - side, you got Defender for cloud, Defender for - containers. - You got the identity side with Entra - where you need Defender for cloud apps, Defender - for identity, ID p twos. This one actually - calls out you need the e five license, - the M365E5. - But I have a lot of clients, and - I say the value of this one, I - have not a lot. I have a few - clients. - They're relatively small. Right? They don't have, like, - a dedicated - security team. They don't have a dedicated SOC,

- but - they're big enough that they are maybe running - e fives, they're running Defender for endpoint, - they're Intune enrolled, - and something gets triggered. Some type of - not necessarily it's a known attack, but something - in Defender is triggered as a potential attack. - Defender is great. It is not the most - intuitive interface - for a IT generalist - to dig through, to figure out if they - need to be worried about a particular - alert or incident or something that is raised

- in Defender. If Microsoft could figure out a - way to really - lower the cost of this, so some of - these SMB - type of - companies - could leverage this to help them with an - investigation of certain alerts and incidents raised in - their tenant. I think there's a lot of - value there in helping - what I I will say more IT generalists - understand - what they should be worried about or concerned - about when different alerts are raised in Defender. - And I think that's where this one fits

- in the blue team is when you actually - have an attack, not taking inputs from the - red team, but actually looking at attacks - and trying to - help people identify - what actually is going on there, helping them - understand the attack path, the replaying the attack, - giving them a summary, - looking through what was the result of the - incident, - what were the different MITRE attack techniques that - we're seeing, like, what are all the different

- components of the attack - to help someone understand that, like I think - that one's there. There's the threat intelligence agent. - Threat - intelligence - agent. I struggle with this one. - Yeah. Why why is that? How is this - different than a secure score? That would be - it in a summary. Threat - Intelligence is - looking at reports - or no. Maybe it wasn't this one. Maybe - it was the green one. No. It's more - the green one. Threat intel - I but I think this is still somewhat - of a

- secure score A little bit. Ish. Very very - much so. So - it's taking - kinda public signals - and public reports and then identifying them out - against in environments - that are out there. So or I guess - your environment. This is where it comes down - to, - like, is it a report at the end - of the day? Like, yes. It's a report. - You get some markdown, - comes out of it, things like that. What - do I go and do with that? Then - it's up to you to to figure it - out. It's interesting what the different reports

- give back. - So, like, if you look at, like, the - attack - investigation agent, for example, like, one of the - primary things that it's intended to do is - give you - a little bit of guidance. So it has - a a classification in the report, like an - incident verdict. So I'm gonna give you a - classification that says, is this a true positive, - a false positive, or benign with transparent reasoning? - So that can be a little helpful. Right? - Like, they drive you a little bit further - down, like, - something that maybe, like, a human doesn't need

- to take a a hard decision on, but - can be human in the loop, things like - that. And that that's a little bit different - than, like, the threat intelligence agent, which is - still gonna give you your executive summary, all - all that kind of stuff. But - it's, - kinda just give you a a a profile - of a potential threat actor - and things that are out there. Interestingly, it - does do a little bit of kinda like - the attack investigation agent or maybe even like - that Rihan agent is doing. It's also kinda

- looking for, like, indicators in your environment. So - if you're thinking about, like, identity logins, things - like that, like, that's looking at IP addresses, - domain names, - e email addresses. - And then - it's trying to leverage, because, again, Defender all - under the hood. Right? You've got this common - core platform in there. Yep. Pull in, like, - enrichment data from Defender. - So if there's, - any kind of known attack vectors, things like - that out there, it'll be able to classify - it against your environment. But to your point, - like, couldn't I just go and maybe look

- at Secure Score or another dashboard? And, like, - that that stuff should be kinda screaming at - me to say go fix it. Maybe this - gives you a little bit more. The other - one that I feel like is kinda similar - to this is you already have and maybe - this is a superset. You already have vulnerability, - the vulnerability assessment, vulnerability management as well that's - in Defender that's already looking at your environment - and telling you where you're vulnerable, where you - have - users, machine software, - settings that are creating vulnerabilities - in your - environment. Again, maybe that's just one of the

- items that feeds into this, but some of - these I feel like it's maybe a little - bit of a - stretch to say we need this agent doing - it because a lot of that functionality - is - already there to a certain extent. But you - kinda, like, want it to go just a - little bit more. Right? Give me the report, - take the output. Like, let me do something. - So so so, like, threat intelligence, for example. - Another one of the outputs that it has, - some of the artifacts. So you get this - markdown report that comes out. It also gives - you a bunch of kinda ready to go

- Kusto queries, KQL queries for your Defender XDR - telemetry. - So that that's cool. Right? Like, not everybody - is a KQL expert, let alone understanding, like, - the schemas as they hop across maybe, like, - a set of tables and log analytics related - to Azure over to your security environment. Like, - that that stuff's, - like, can be weird to poke at. And - unless you, like, really know the schemas or - have internalized them, like, I could see that - being super helpful, but, like, why can't you - just go run those queries for me? Like, - so great. So you gave me a bunch - of ready to use Kusto queries that are

- scoped to my environment that are gonna help - me search for indicators - of threats in my environment. Why don't I - just go run like, keep running it. Right? - Keep running it in a loop, - come back, and then kinda give me the - outputs, and let's go from there. So you - can see, like, the potential. Like, I think - this is where they would go with it, - but who knows? It's expensive to run it - in a loop, Scott. - That's gonna cost me both arms and legs - instead of just one arm and a leg. - Yeah. - But I do agree. Like, it would be - good to see it a little bit more

- proactive, - automated - type of stuff. The other one are you - ready to move on to the detection authoring - agent? This is another one that intrigues me. - This one, I would say I can also - see a little more value here - of - this one is - to help you build - exactly what that sounds, detections. - What type of logic do you need to - have? What type of rules do you need - to have in place? - How should you be looking at your data - to help create

- detections in your environment? - And - this one, to your point earlier about some - of the KQL and the schemas and generating - that, a lot of detection alerts right now - are based off - of KQL queries that you write or you - get from content packs and Sentinel where you - have all your data going in there, but - you need to write your - KQL queries and build your I completely am - blanking on the name of what they are - right now in Sentinel, but essentially your automated - triggers. It sees this data. It runs this

- workbook every five minutes, fifteen minutes, hour, - but it still requires you to author all - of that. - This one - is - designed really to help people create - KQL based - detections, - map - detection coverage to MITRE attacks, - tune their detection methods. Especially your alerting logic. - Right. It's helping you build your alerting logic - in a way that maybe you don't need - to go in and understand all of that - logic to the

- level of depth that you need to now. - I look at this kinda like how I - use AI to write PowerShell scripts. I can - go write a PowerShell script. - Sometimes it's easier just to let AI figure - out the logic and the tweaks and the - little nuances of it. I think this is - the same way. You can still go write - KQL. You can go write all of this. - This just helps with some of the nuances - and helps you refine - how you detect - or how you detect more alerts, - potential compromises in your environment, which I think - can be a good thing.

- It can be very nuanced to write these - things - where you don't give so many false positives - that you just start ignoring them. I think - you're still a little and that's where it's - human in the loop. Right? Like, you're - depending on those - and so the agents, because it's a funnel, - they're meant to they are meant to feed - into each other. So but you you're sitting - there coming out of that threat intelligence agent, - which had those classifications, - like false - positive, true or or, yeah, false positive, true - positive, benign, and then you're taking it and - kind of passing it through passing parts of

- it through, I guess, because you're kinda left - with, like, - the threat intelligence agent seems like it's giving - you a KQL to go and look and - hunt a little bit for yourself. - And then something like the detection authoring agent - is coming in. It's taking that - it's taking that report, and then it's still - generating you yet another set of - alerting - and detection logic that then you would go - and - deploy in your environment if you so choose. - Because, again, there's gonna be more, like, rationalization - here for you to say, hey. - Is this thing that the agent's telling me

- something that I actually wanna go do? And - then that would make it very secure. Right? - You're sitting out there looking at a number - of, like, the what what I have to - do now? Like, a is still at 800 - or I don't even know. I haven't paid - attention to, like, what the boundary is. But, - like, you're looking at a number, right, that - really doesn't have any meaning, and then you - gotta take a step back and figure out. - So, hopefully, some of these reports - come with them come with some of that - stuff. I would have loved to have seen, - like, sample reports in the documentation or something - like that, but they didn't have any of - that. They just give you the schemas, the - reports, or the high levels. What's gonna come

- out of it? And then the final one, - green team. Ready to move on to green - team? The posture prioritization - agent. This is the one I was thinking - of that I'm like, this is just the - agentic agentic version - of the Secure Score. This helps teams and - stakeholders - understand what to fix first and why. In - my mind, that was the whole point of - the Secure Score. Like, let me go look - through the Secure Score and find where I'm - the most vulnerable or where I can get - the most bang for my buck in terms - of fixing and use the Secure Score to - prioritize it. Maybe this takes it a little

- step further. It's looking at the inputs of - where are - attacks being seen, - what is - some of the - attack investigation - or the threat intelligence, - the detection, - authoring that you're writing. - Is this taking inputs from the blue team - to kinda reprioritize? - Because the one thing I would say about - Secure Score is - it was very much, is this configuration on, - and how many users, how many machines, etcetera, - is it applied to? You just get a

- number. The prioritization - of Secure Score - wasn't - necessarily based - on what I would say real world data, - real world analytics, and what's going on in - your tenant other than a settings offer on - across x number of people. So maybe this - helps with a little bit more real time - prioritization, - taking in a few more signals other than - just is a configuration on or off. But - this is going to not go in and - make fixes for you, going back to your

- human in the loop, the read only, it's - just going to tell you here's the order - you should go apply all these settings or - fix all these things in your environment. The - other thing we should maybe talk about here - as we're kinda getting towards the end is - orchestrating these things or pushing them together. It's - great. So you you get all these agents - that are out there. - Like, what does it look like - to run them? I think the the - the most interesting one to me was probably - the playbook things or the playbook play. Go - go and kick these things off. So if

- you did have something in the Defender portal - pop up, so you get an incident to - where you get an alert that pops up, - you can go run these playbooks, and then - the playbooks act as orchestrators - for the underlying agent definitions that are out - there. So they can kinda - go start the flywheel for you, but then - you'll still be human in the loop for - kinda those decisions along the way that we - talked about. Yep. And these are very much - this is something I would love to be - able to start playing with. Maybe - once it becomes generally available, I'll just spin - it up for a few days to play

- with it. But Playbooks very much came from - Security Copilot. Like, people already have - Security Playbooks in there running a series Yeah. - I mean, you can see the evolution here. - Like, this is just I feel like this - is even further integrating Security Copilot into where - people work. Now you don't have to go - over there, you can go trader these playbooks, - tie in these agents that, I mean, realistically, - maybe are a part of Security Copilot or - this is the next - iteration of Security Copilot - within your environment. But, yeah, the playbooks look

- like they're gonna be that main - that main service that you're gonna go use - or that main functionality - that you're gonna use to go trigger all - these prompts and these agents from various areas - within Defender or within your environment. Defender then - it's interesting. It becomes part of that single - pane of glass as well, which it was - meant to be anyway. So - you you can kinda come at it a - couple different ways. Exactly. - Anything else? Sessions? I mean, sessions are just - monitoring your agent work. Session, just like any

- session in AI. But it's interesting. I'm curious - to see where this goes. I am curious - to get my hands on it and - use it sparingly - so I don't run out of money. But - and to see to your point where agents - are going, this is still very much a - lot of read only. - As - these get better and better, is there a - point we get to - where an alert gets detected - in one of the, I guess, the blue - team agents would be detecting them and then - triggering off a green team agent where it

- actually can go in and - make the resolution, make the fix to whatever - the issue is? We were talking about it - earlier. I feel like I'd still be a - little nervous of that. There's a whole weird - thing here. Right? So so do you trust - these identities? So so, like, that recon agent - is out there, but it's it's got a - bunch of those graph permissions. Like, okay. Hey. - And it's got Azure subscription scope rules in - it. So, sure, they're read only, but it - does expose your configuration. - Like, - what what happens if that configuration, - like, report doesn't get published to the right

- place? - There's a big question about who supervises these - things. - So if you're gonna be, like, human in - the loop for these things - and you're gonna be supervising it, I think - you'd want those folks to have kinda the - same or greater permissions than the agent. Right? - You wouldn't want the agent to be able - to run away past that? Yeah. Because you - wanna be able to go in and validate - some of the data coming back, I think. - I would. But what does that look like - in a world of, - like, junior to senior level, like, - folks in your security operations center. Like, if

- you're sitting out there, typically, like, junior folks - are gonna be, like, very scoped down in - their permission. Like, now you can't have them - fully monitor the agent because, one, they could - be exposed to configuration they're not supposed to, - and, two, they don't have the same rights - as as they need to anyway. - And then I think it will be interesting - as more, like, activity based things come in. - So where you can maybe do things like, - oh, say, like, disable a user account based - on, on, like, anomalous - logins or things like that today. - Where do you start to kinda peel back

- permissions - or become more permissive, I guess, in in - what you put out there? And when do - you start to kinda tear down approval gates? - Because if you go through the documentation, - one of the things that'll pop up a - couple times is - the suggestion, - not like the discrete guidance to go do - this, but it is definitely a suggestion - to just reduce, like, approval gates and some - of the friction that's there for those agents. - So, like, hey. It's proven. Like, I I - had to do this thing. It feels like - it's deterministic, and it's out there. Is it

- gonna be truly deterministic? - Right. Who knows? Maybe it is today, but - is it when, like, the next underlying, like, - foundation model comes out or things like that? - Oh, that that's the other one to think - about is I didn't see any place where - it's explicitly - called out - what - models - these agents use under the hood. It's kinda - hand wavy. It talks about, like, - OpenAI - Foundry models, Azure OpenAI Foundry models, - but which ones and what's out there. Because

- very much, like, sometimes, like, the model does - have a big play - in - what type of output you're going to get - from the same set of inputs. Right? Like, - you'd be as structured as you wanna be - in that agent configuration - and and flow of it. But throwing it - against, like, GPT five five versus - the new five six, like, Sol Luna Terra - stuff, like, gives you very different results. So - my understanding of this - based on the blog post, - and I didn't see it in the documentation,

- is that it is very much a - multimodel - where it might be using some OpenAI, - it might be using some Claude. - It also looks like it is using and - we didn't even talk about this, the MAI. - I don't know how you pronounce Microsoft's models - now. May, my, the CyberOne - Flash, which is Microsoft's - new security model - that they claim it scored better above Mythos. - I think all of these companies run their - own benchmarks - on their own models. Your mileage is gonna

- vary with benchmarks and what are out there. - Yeah. On - how it is, but it does say it - oh, here. It does. Project perception will take - advantage of that Cyber One Flash model from - Microsoft that they created. For which parts? For - what? For which parts? It doesn't say I - don't know. I don't know. I don't know. - I don't know. I don't know. I don't - know. I don't know. I don't know. I - don't know. I like my kids. Ask me - impossible questions. Where all your SCU cost is - coming from. It's all hidden away under the - hood. That's - my understanding is I don't know that they're - publishing the model because I mean, do do

- they even know which one gets passed off - to which model, or is this one of - those - where there's, like, the initial - foundational - model that initially prompts it, and then the - agents go out to find the right model - based on the tasks they wanna do? There - is some blurriness - there in which models it's using where. Other - than that, it is gonna use Microsoft's model, - its - aspects of it. How that propagates out to - other ones - is absolutely - vague.

- I think it's gonna take some time to - figure it all out. Yes. Yeah. No single - model will be optimal for every security task. - Cyber defense requires the right model for the - right problem at the right time. - Project perception, the right model is determined by - the combination of quality, reliability, latency, and cost - rather than relying on a single model, blah - blah blah blah blah. So it is using - a slew of models that it picks based - on some rules Microsoft has in place. Yeah. - Got it. Interesting. But we will have that, - the blog post. - So there's the blog post out there. There's

- the Microsoft - web web page out there that we'll have - in the show notes, and then - all the learned documentation - is already available. So you can go read - all about this, - see everything, you just can't get access to - it yet. They do have a button for - contact sales - where you would normally - go to, like, buy a preview or get - started with it. Anytime a website tells me - to contact sales, I think of this isn't - gonna be cheap. - That's just - Yeah. My pre programming. - So I'm excited to get my hands on

- it and try it out and see what's - in there at some point in time. I - don't know when that will be. We'll have - a follow-up episode when we can actually play - with it. Follow-up. - Ben Ben plays with agents in the clap. - Yeah. No. I think it'll it'll be interesting - to see how it forms - over time and, well, how, like, how the - manifestations - of those agents - change. Like I said, they're very they're very - high level today, - and it's gonna be it's gonna be quite - the journey to see them go from being - able to affect, like, tenant configuration changes, things

- like that. It's like a hard no and - read only to - more broad yeses and what can be out - there and be done with it. With that, - I have a meeting coming up. I would - love to hear other people's thoughts on this - if somebody listening actually - is in the public preview and has access - to it and you're allowed to talk about - it. Feel free to reach out. Like I - said, this past week, we had some discussion - on Reddit about some of these things on - the MS Cloud IT Pro - subreddit. Feel free to discuss on LinkedIn as - well. Would love to - chat with people there and get people's thoughts.

- Also, Scott, if there is anybody that's gonna - be at TechCon three sixty five in Seattle - the end of August, - it'll be, like, two weeks after this episode - comes out. I will be out there doing - a few sessions and would love to connect - with anybody that will be out in - Seattle at the end of August for that - conference. Got any codes or anything for them? - I do have a code. I just I - should know what it is. Oh, I shoulda - looked it up. It is - it can be in the show notes. I - think it's I'm not going to even - make one up. Kinda put you on the

- spot, so we'll put it in the show - notes. There's two Bens you did. We'll put - it in the show notes. There's two Bens, - and I can't remember if I'm Ben or - Ben too. I mean, you are well, you - are Ben as well. That's okay. I am - Ben as well, and I was trying to - find one of my posts here that - oh, here it is. I know where I - can find it. Oh, super quick. How quick - can I go? I found a post that - has my code. - Ben200BEN, - and then the number 200. We'll save you - $200. In February. And TekCon in Seattle? TekCon

- in Seattle. Yep. August - '28. - Alright. Well, thanks for - sitting on. It's always good to walk through - a spec sheet with you here, Ben. Yes. - For sure. Thank you. Thank you for your - perspective on perception. Perception. Okay. Your perception of - perception. We're we're gonna get all we're gonna - get all twisted around. - Alright. It's Monday. Go enjoy your meetings. Thanks, - Ben. Alright. Thanks, Scott. - If you enjoyed the podcast, go leave us - a five star rating in iTunes. It helps - to get the word out so more IT - pros can learn about Office three sixty five

- in and Azure. - If you have any questions you want us - to address on the show, or feedback about - the show, feel free to reach out via - our website, Twitter, or Facebook. - Thanks again for listening, and have a great - day.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Episode 441: Microsoft Ignite 2026: Autopilot, Workspaces, and Spot Plus

    Nov 19, 202648 minAudio

    Welcome to Episode 441 of the Microsoft Cloud IT Pro Podcast. In this episode, Ben and Scott take a first pass at Microsoft Ignite 2026, starting with a question a lot of IT pros are asking: is the pace of change...

    Credit the people on this episode

  2. Episode 440: The J Factor: Bringing Your Experience to Generative AI

    Nov 5, 202638 minAudio

    Welcome to Episode 440 of the Microsoft Cloud IT Pro Podcast. Recorded live at the North American Cloud and Collaboration Summit in Branson, Missouri, this MS Cloud on the Road conversation has Ben sitting down with Jay...

    Credit the people on this episode

  3. Episode 439: Azure SRE Agent vs. the Azure Copilot Troubleshooting Agent

    Oct 22, 202638 minAudio

    Welcome to Episode 439 of the Microsoft Cloud IT Pro Podcast. In this episode, Ben and Scott dig into two Azure agents with confusingly different names: the Azure Copilot Troubleshooting Agent and the Azure SRE Agent....

    Credit the people on this episode

  4. Episode 438: Taking your WorkOS and LLM Wiki to the Next Level

    Oct 8, 202647 minAudio

    Welcome to Episode 438 of the Microsoft Cloud IT Pro Podcast. Back in Episode 429, Ben and Scott introduced LLM wikis. In this episode, they come back a few months later to compare what those wikis turned into after...

    Credit the people on this episode

  5. Episode 437: Nasuni and AI-Ready Unstructured Data with Nick Burling

    Sep 24, 202640 minAudio

    Welcome to Episode 437 of the Microsoft Cloud IT Pro Podcast. In this episode, Nick Burling, Chief Product Officer at Nasuni, joins Ben and Scott to talk about getting unstructured file data ready for AI. Nick digs into...

    Credit the people on this episode