KOPYA NG SALIN 🔗
WEBVTT
00:00:00.560 --> 00:00:03.600
Hello, welcome to another edition of Third Party Threateners Podcast.
00:00:03.759 --> 00:00:08.000
Today I have a special guest, Michael Berman, who's a CEO for end contracts.
00:00:08.080 --> 00:00:13.519
And I'm going to let Michael do his intro about his background and expertise, and then we'll get into the five questions as we normally do.
00:00:13.679 --> 00:00:14.000
Michael?
00:00:14.320 --> 00:00:16.000
Well, Greg, thank you so much for having me.
00:00:16.160 --> 00:00:17.519
Excited to chat with everyone.
00:00:17.679 --> 00:00:19.120
My name's Michael Berman.
00:00:19.199 --> 00:00:22.239
I'm a recovering attorney, turned entrepreneur.
00:00:22.559 --> 00:00:26.000
So my background and my approach to this is a little different.
00:00:26.239 --> 00:00:38.399
I've had the privilege of leading in contracts, which started as a third-party risk software and services firm now for 17 years and have seen a lot of changes, certainly during that time.
00:00:38.560 --> 00:00:44.399
And I have the responsibility for dealing with third-party risk when I was general counsel and practicing law.
00:00:44.640 --> 00:00:53.280
So a lot of different perspectives, but the same goal that all of us who are interested in third-party risk have, which is how do we mitigate this risk?
00:00:53.679 --> 00:00:54.159
Exactly.
00:00:54.399 --> 00:00:54.880
That's great.
00:00:54.960 --> 00:00:55.200
Thanks.
00:00:55.280 --> 00:00:55.920
Thanks, Michael.
00:00:56.000 --> 00:00:59.759
So one did the five questions that we asked to get to our guests a little bit better.
00:00:59.840 --> 00:01:01.280
The first one is your morning routine.
00:01:01.439 --> 00:01:06.959
When you're not managing growth at end contracts or writing books, which I, by the way, plug your book on Amazon, check it out.
00:01:07.519 --> 00:01:09.519
It's the upside of third-party growth.
00:01:09.760 --> 00:01:10.319
Upside of third contracts.
00:01:10.400 --> 00:01:10.799
Yeah, that's right.
00:01:11.760 --> 00:01:13.200
I'll make sure I got a plug in there for you.
00:01:13.599 --> 00:01:15.040
Just live in fear of it.
00:01:15.120 --> 00:01:19.519
And the reality is if you can conquer it, your organization's actually going to perform a lot better.
00:01:19.920 --> 00:01:20.560
Right, exactly.
00:01:20.719 --> 00:01:26.159
When you're not managing growth at end contracts or writing the books, uh, how do you start a regular sort of non-work morning?
00:01:26.879 --> 00:01:30.959
I am a big at this point in my life, a fitness enthusiast.
00:01:31.040 --> 00:01:41.280
So I found I really love to start my day with some type of cardio activity, uh, both so I can stay live as well as it is wonderful stress relief.
00:01:41.519 --> 00:01:44.560
Yeah, very same here, same here, Michael.
00:01:44.640 --> 00:01:47.840
Uh okay, so we start off with the hook.
00:01:47.920 --> 00:01:51.200
Uh just a five-minute quick talk about something that's interesting about timely.
00:01:51.359 --> 00:01:57.920
So, Michael, in your book about talk about transforming risk and compliance from a defensive burden into a strategic advantage, very similar to my thought process.
00:01:58.000 --> 00:02:04.400
But in financial services, most third-party risk management is still driven by regulatory compliance, which is checklists and annual vendor assessments.
00:02:04.480 --> 00:02:12.400
Are banks and credit unions actually reducing their vendor risk surface today, or are they mostly just satisfying the examiners and moving stuff around on the deck of the ship, as it were?
00:02:12.639 --> 00:02:13.280
Loaded question.
00:02:13.520 --> 00:02:15.680
I would tell you it's getting better, correct?
00:02:15.840 --> 00:02:21.280
It's getting better in that banks and credit unions, of course, and we take this way back in time.
00:02:21.599 --> 00:02:26.719
The first guidance really is uh in the late 90s and then early 2000s.
00:02:26.879 --> 00:02:32.960
And like a lot of regulations enforced at the largest institutions first and then moving on down.
00:02:33.199 --> 00:02:50.960
And so over time, though, as a third-party risk management's matured, people are viewing it more and more seriously, especially as they see the real-world impacts when a vendor, cybersecurity vendor has a problem, so fourth-party vendor, and all of a sudden now they've got a problem.
00:02:51.120 --> 00:02:54.800
And the list of these things just continues to unfold.
00:02:54.960 --> 00:02:59.520
Probably one of my favorite recent stories is a company called Dwell.
00:02:59.680 --> 00:03:10.240
So Dwell is a startup AI company funded by Y Combinator and Insight Partners, and uh had allegedly would read SOC documents for you.
00:03:10.319 --> 00:03:14.240
And a whistleblower came forward and said it was fraudulent.
00:03:14.479 --> 00:03:20.319
And turns out the exact same summary was generated for every SOC report they ever had.
00:03:20.479 --> 00:03:22.879
And the funders pulled out and all that happened.
00:03:22.960 --> 00:03:24.560
And of course, all of this is alleged.
00:03:24.639 --> 00:03:29.280
Um, but the fact that they lost all their funding is kind of a bit damaging.
00:03:29.520 --> 00:03:46.319
But it goes to the point of people are paying far more attention now because we live in a world where you've got to trust but verify all of these things, which includes the background of the vendor and all the things that the vendor does to make sure you get accurate results.
00:03:46.479 --> 00:03:55.599
So, to answer your question directly, it's evolving from a compliance exercise into one where people are taking it far more seriously.
00:03:56.319 --> 00:04:03.680
My last favorite analogy, think about if you're working in a whatever organization you're working in, how much time you spend reviewing people.
00:04:03.919 --> 00:04:06.000
That's right, HR reviewing people.
00:04:06.159 --> 00:04:11.520
Can you imagine not reviewing your people at
Load full transcript
all during a year or only talking to them once a year?
00:04:11.759 --> 00:04:17.519
Well, your vendors are have access probably to more data and create more threats than your people do.
00:04:17.759 --> 00:04:22.959
You should at least spend as much time as you spend looking at your people, looking at your vendors.
00:04:23.120 --> 00:04:26.240
And I think, again, organizations are starting to really understand that.
00:04:26.560 --> 00:04:30.160
I'm totally going to steal that uh analogy, Michael, the the review your people.
00:04:30.480 --> 00:04:33.759
Because that that totally that that hits home for most folks.
00:04:33.839 --> 00:04:37.920
So would you would you, as a manager or leader, not review your folks just once a year?
00:04:38.079 --> 00:04:39.040
Would you not talk to them?
00:04:39.199 --> 00:04:41.439
You at least do mid-year check-ins, right?
00:04:41.519 --> 00:04:42.959
You know, at a minimum.
00:04:43.040 --> 00:04:45.839
Um, and if you're not, then you're not a very good manager, to be honest with you.
00:04:46.000 --> 00:04:47.600
So it's a different story.
00:04:47.759 --> 00:04:49.360
It's a different podcast.
00:04:49.519 --> 00:04:49.839
All right.
00:04:49.920 --> 00:04:54.639
So uh big question number one is moving beyond uh moving beyond static compliance to threat?
00:04:55.839 --> 00:05:06.000
As a financial institution face progressive interagency guidance and third-party relationships, the interagency guidance, I think a credit union uh regulatory is starting to look uh more heavily into the space as well.
00:05:06.160 --> 00:05:09.680
I've seen uh increased increased scrutiny from some of my clients in the credit union space.
00:05:09.839 --> 00:05:13.279
Yeah, threat actors target vendors in real time between annual assessment cycles.
00:05:13.360 --> 00:05:16.720
So, how or how do organizations align static regulatory compliance?
00:05:16.879 --> 00:05:23.680
How do they build off regulatory compliance requirements for dynamic continuous threat haunting and active telemetry monitoring without overwhelming their compliance teams?
00:05:24.079 --> 00:05:24.560
Great question.
00:05:24.720 --> 00:05:32.319
And first, there is a real gap that you've very accurately identified your work as well.
00:05:32.639 --> 00:05:34.560
And that's what this exists.
00:05:34.879 --> 00:05:44.800
So most third-party programs are built around asking a question: was this vendor secure on the day I onboarded or the day I did an assessment?
00:05:45.040 --> 00:05:47.759
And there's value there, and I'll get to that in a minute.
00:05:47.920 --> 00:05:57.120
But what threat hunters are asking is a totally different question, which is also a great question, is anyone inside my environment right now taking my data?
00:05:57.360 --> 00:06:00.560
Is something going on now that I need to address?
00:06:00.800 --> 00:06:05.519
And so both are legitimate questions coming at a problem from different angles.
00:06:05.680 --> 00:06:08.800
And
unfortunately, the answer is we got to do both really well.
00:06:08.959 --> 00:06:10.720
So I'll give a different analogy.
00:06:10.959 --> 00:06:14.800
When I'm doing my risk assessment, there are things I can learn.
00:06:14.959 --> 00:06:18.160
Think of it as I'm getting an x-ray of my vendor.
00:06:18.240 --> 00:06:26.560
I'm gonna understand their bone structure, I'm gonna understand so much data about how they're financially structured, a contract, all these wonderful things.
00:06:26.800 --> 00:06:27.759
Is that important?
00:06:28.160 --> 00:06:29.040
Sure it is.
00:06:29.199 --> 00:06:36.800
But then when I'm doing my threat hunting, I get a live video feed of what's going on right now inside their environment.
00:06:36.959 --> 00:06:38.399
That's also really important.
00:06:39.120 --> 00:06:44.720
But yet you're gonna have different threats, and I have run into organizations that ignore it.
00:06:44.800 --> 00:06:59.839
I've run to organizations where all they care about is cybersecurity, which sounds great, and they've invested all these resources until the vendor fails financially, or until the vendor has a business continuity issue or a compliance issue, because there are other risks beyond cyber.
00:07:00.000 --> 00:07:12.560
That's not diminishing in any way the cybersecurity threats, which are extremely real, but they intertwine because if a company is financially in trouble, they're going to cut back on their cybersecurity.
00:07:12.639 --> 00:07:14.399
They can't afford it anymore.
00:07:14.639 --> 00:07:20.240
So we've got to address that gap failure of thinking one side has it right.
00:07:20.399 --> 00:07:35.199
They interact together, and the GRC and security operations need to grow together, even if you came up from different sides of that equation, because maturity and risk ratings matter and threat vulnerabilities matter.
00:07:35.360 --> 00:07:39.759
So we've got to kind of bridge that gap so that we can make it matter.
00:07:39.920 --> 00:07:52.240
So things like solar winds, move it, change healthcare, all of these kinds of issues that people live through from a vendor breach perspective start with pristine vendor reviews.
00:07:52.399 --> 00:07:56.319
So the threat is real because people miss things.
00:07:56.800 --> 00:07:59.360
And no organization wants to have that missed.
00:07:59.519 --> 00:08:04.319
So we've got to get a better handle of how do we bridge the gap between those two things.
00:08:04.639 --> 00:08:14.319
So the core issue here is how do we translate that translate this information into action instead of just more data?
00:08:14.560 --> 00:08:18.079
Because that's where most third-party risk programs get lost.
00:08:18.319 --> 00:08:19.680
Hey, I'm gonna subscribe to this.
00:08:19.759 --> 00:08:48.159
Now I got more data, now I'm gonna subscribe to, and you can get an endless amount of and wind up in information overload relatively quickly, because it's how you interpret that data and what actions
you take, and that's your framework that third-party risk specialists thrive on so that we can create practical mechanisms that will work, that will allow us to mitigate that risk as we move forward, and you wind up with that balancing act.
00:08:48.399 --> 00:08:55.039
So you might have 800 vendors, okay, but that doesn't mean I need to do threat monitoring for 800 vendors.
00:08:55.279 --> 00:08:59.360
There are certain vendors where it's gonna matter a lot, and others where it's not gonna matter.
00:08:59.440 --> 00:09:12.480
So that's that shared taxonomy between your threat team and your vendor kind of traditional compliance team, you gotta have that shared so that you're all working off the same threads.
00:09:12.639 --> 00:09:19.519
Because if cyber team thinks something's really critical and your compliance team says it's irrelevant, there's a gap.
00:09:19.600 --> 00:09:31.200
And we got to solve that gap because we all work for the same company and we want to make sure we've got so we've got to map that risk in the appropriate way to really allow us to have that.
00:09:31.360 --> 00:09:33.440
And then governance.
00:09:33.600 --> 00:09:46.000
So we've got to change what the risk committee sees instead of going, we're almost done with our risk assessment, to how do we incorporate kind of a real-time threat vulnerability into that assessment?
00:09:46.159 --> 00:09:56.240
And the assessment's not a once-a-year activity, because if these vendors are critical, we need to how do we monitor it on a monthly basis, on a weekly basis, even daily basis?
00:09:56.399 --> 00:09:59.120
So if something has gone wrong, we get an alert.
00:09:59.200 --> 00:10:01.759
So I'll go back to my example, the Movit.
00:10:02.240 --> 00:10:07.919
Well, if all of a sudden you have a sophisticated enough system to go, hey, Movit's had this issue.
00:10:08.159 --> 00:10:10.639
Which of my vendors are affected by Movit?
00:10:10.879 --> 00:10:14.879
Well, if you've mapped things appropriately, you can have that level of transparency.
00:10:14.960 --> 00:10:23.519
So you're now in Vinzer incident response, which allows you in that program of action versus, I don't know.
00:10:23.679 --> 00:10:28.480
Let me go survey my vendors for the very first time to find out which ones may have been affected.
00:10:28.720 --> 00:10:35.519
Which is what a lot of organizations did when Movit first came out, which drove a lot of folks just absolutely nuts because it's like asking your vendor, are you breached?
00:10:35.759 --> 00:10:39.840
If the vendor's been breached, they're probably not going to tell you over an email in that way.
00:10:39.919 --> 00:10:42.799
They'll probably send you a letter from their legal attorneys, right?
00:10:42.879 --> 00:10:45.440
When it gets to that point, usually the attorneys get engaged, right?
00:10:45.519 --> 00:10:47.759
It's not just a nice email that says, Yeah, we've been breached.
00:10:47.840 --> 00:10:48.799
We'll contact you next week.
00:10:48.879 --> 00:10:49.039
Right.
00:10:49.279 --> 00:10:50.879
Same thing with the movie, countless ones.
00:10:50.960 --> 00:10:56.960
And surprise, surprise, nobody ever said yes unless they'd already told us that through a legal attorney that they've been affected by it.
00:10:57.360 --> 00:10:59.120
This is where the contract helps.
00:10:59.279 --> 00:11:16.080
So knowing that they're going to disclose fourth party vendors, requiring the disclosure of these fourth party vendors is highly relevant because, Greg, to your point, if they already have a breach or know of it, the vendor is in this position of, well, we use Move It, but I don't know if my data has been breached or not.
00:11:16.240 --> 00:11:17.679
Am I part of the Move It Breach?
00:11:17.759 --> 00:11:19.120
Am I not part of the Move It breach?
00:11:19.360 --> 00:11:23.279
What of my files was my client affected by this or not?
00:11:23.440 --> 00:11:27.200
And those are all really legitimate questions the vendor can ask.
00:11:27.360 --> 00:11:31.440
But as the client, I'm left in the dark completely by this.
00:11:31.600 --> 00:11:48.240
So this is where all that preparation, frameworks, assessments, really get your practical implications are in place of how do I get that so I can govern things and address those risks in an operational way so that we can take action.
00:11:48.480 --> 00:11:54.159
And things like AI in particular is making all this far more complicated.
00:11:54.399 --> 00:12:05.679
And so it is more important than ever to have good third-party risk governance in place because it's now going to get a lot more complicated.
00:12:05.840 --> 00:12:15.120
If you thought the cloud made third-party risk more complicated, AI has upped it by a hundred times because of the way this technology works.
00:12:15.360 --> 00:12:26.879
And so it's now becoming critical to have this bridge between cyber threat huntering and traditional GRC, because you're not going to have a way to govern this.
00:12:27.120 --> 00:12:53.200
And instead, you're just assuming the risk, in which case you should brush up on your cyber insurance because it's a question of um and so those are those are some of the areas where I think we can evolve to kind of bridge that gap to have a real solid foundation for managing third-party risk that involves both threat hunting and traditional risk and compliance to solve the holistic risk.
00:12:55.440 --> 00:12:57.759
Fourth party risk and shadow tech AI sprawl.
00:12:58.240 --> 00:13:05.360
Michael, with the massive expansion of fintech integrations, cloud infrastructure, and third-party AI tools, vendor ecosystems are becoming hyper-connected webs.
00:13:05.519 --> 00:13:07.759
Not that they already weren't, but they're just even worse.
00:13:08.000 --> 00:13:14.159
But from an integrated risk management perspective, how would risk leaders handle fourth-party exposure and shadow AI before it turns into a major breach?
00:13:14.320 --> 00:13:18.000
It does talk about now the fact that most of the AI is nth party to us.
00:13:18.159 --> 00:13:20.960
We don't have a direct connection with anthropic or all those others.
00:13:21.039 --> 00:13:25.919
We we talk to the the companies making the software who's consuming the AI tool.
00:13:26.240 --> 00:13:31.840
So this is a great question, and one that is key to practitioners and third-party risk.
00:13:32.399 --> 00:13:39.039
Because the problem is the perimeter of governance, because of AI, has just dissolved.
00:13:39.200 --> 00:13:40.320
What do I mean by that?
00:13:40.559 --> 00:13:47.679
So traditional GRC is built on the assumption that we can have a we can know what technology our firm is using.
00:13:47.840 --> 00:13:58.159
In the AI world and with nth-party vendors, as you just described, I may not be able to know what technology all of my data might touch.
00:13:58.480 --> 00:14:00.639
It is, we're living in that world now.
00:14:00.799 --> 00:14:23.279
So, and of course, shadow AI makes it worse because an employee can adopt an AI tool, work in a work email, product team wires up SaaS integration to something, and then all of a sudden my data through an API key is going, and no procurement team was ever contacted, no security team was contacted, and uh, we're in a world of high risk.
00:14:23.519 --> 00:14:32.879
And so your vendor inventory is probably incomplete, and the gap will continue to grow as AI continues to expand.
00:14:33.039 --> 00:14:35.360
So that's the problem that we face.
00:14:35.600 --> 00:14:49.840
And so AI supply chains, as Greg's already articulated to some extent, break the traditional vendor management because when a vendor adopts AI, they are typically leveraging a large language model.
00:14:50.080 --> 00:15:01.120
And those large language models, and they may enhance it with other things, they may even enhance it with a small language model, they may enhance it with various data that's unique to them.
00:15:01.440 --> 00:15:23.279
I'm trying not to use all the technical terms because I don't want to turn this into AI vocabulary, but there's a lot that they can provide, but you're adopting the vendors' AI foundational model provider, you're adopting their cloud infrastructure, you're adopting their training data pipeline and any subprocessor they have underneath.
00:15:23.840 --> 00:15:29.279
And the idea that also as they continue to build and evolve, that's going to change.
00:15:29.440 --> 00:15:31.519
And so, how do I keep track of all of that?
00:15:31.679 --> 00:15:37.440
And the answer is your vendor can't even explain right now their own supply chain.
00:15:37.600 --> 00:15:45.679
So you're asking questions the vendor really probably doesn't know how to answer unless you get their their heads of engineering on the phone.
00:15:45.840 --> 00:15:55.360
And even if they could explain, it's a point in time that's changing on an ongoing basis as Anthropic launches a new model or uh Chat GPT launches a new
model.
00:15:55.600 --> 00:16:01.600
So the traditional diligence and a questionnaire, we're gonna have to rethink how we do that.
00:16:01.840 --> 00:16:03.919
All hope is not lost, I assure you.
00:16:04.159 --> 00:16:16.480
So we know the risk behaves differently, and we know that we've got all these fourth parties, and again, your CRM, your HR tools, they may be adopting AI, and so we've got to keep up with that as too.
00:16:16.639 --> 00:16:25.120
So we need to expand the unit of governance from just the vendor to data flows, wherever that data flows.
00:16:25.360 --> 00:16:32.240
So that's a huge shift because we need to both govern companies, because they have risks, and we need to govern data.
00:16:32.320 --> 00:16:46.240
So it's not an either or, and we need to understand that the contract now creates a risk framework that we can leverage as a one point in time to help us with these things as we move forward.
00:16:46.399 --> 00:16:57.919
Because there are still vendors out there, we just need to make sure we are going to hold them accountable for the decisions they make and understand the data flows so we know which vendor we can hold accountable for those things.
00:16:58.240 --> 00:17:05.359
So we have to have both who are our vendors, what systems does our customer data touch, who gets to make those decisions.
00:17:05.599 --> 00:17:14.640
If you go down the path of a GenTIC AI where the AI is making actual decisions, we also then need to understand who controls that and where the data flows.
00:17:14.799 --> 00:17:27.279
And then we need to build discovery questions, the questions you might think of for onboarding a vendor as continuous discussions, so that we understand how that discussion can work.
00:17:27.599 --> 00:17:30.880
And the the real kicker is a sanctioned path.
00:17:31.039 --> 00:17:35.119
Meaning if someone starts to violate it, how do we cut them off immediately?
00:17:35.920 --> 00:17:42.799
And that that's new to people, especially in the financial world, because you're used to letting your vendor know, hey, this isn't gonna work.
00:17:42.960 --> 00:17:44.640
Well, now your data's at risk.
00:17:44.720 --> 00:17:46.400
So you don't have that luxury.
00:17:46.480 --> 00:18:02.799
You've got to have a clear escalation path that allows you to cut it off the moment you have issues, so that you have a clear enforcement path as you move forward, so that when discovery generates a long list, it's not creating a new work stream.
00:18:03.039 --> 00:18:07.839
You have a very clear path of what to do so that you can retool.
00:18:07.920 --> 00:18:21.039
And you're gonna, it's a cultural change because the old ways we do procurement are going to go out the door to this new path of think of it like a malicious website you found, you gotta cut it off right then.
00:18:21.200 --> 00:18:27.200
And you can always turn it back on, but you gotta be able to interrupt the work because of you have that.
00:18:27.359 --> 00:18:30.880
Now, when it comes to like, well, what questions do I ask?
00:18:30.960 --> 00:18:35.839
What data do I can I even gather from these vendors that are leveraging AI?
00:18:36.160 --> 00:18:41.839
The reality is an assessment that explains their model isn't going to work very well.
00:18:42.319 --> 00:18:45.680
Asking these big narrative questions, we need certificates.
00:18:45.839 --> 00:18:56.640
So things like a SOC 2, an ISO 42001, model evidence, model verification evidence, documented data retention, documented data practices.
00:18:56.880 --> 00:19:06.160
Well, I've been training folks to say, uh, hey, if you go to your vendor and they they have AI adopted and they don't have a framework for risk management around AI, that's a problem.
00:19:06.319 --> 00:19:11.920
But if they have a risk management framework, then you can ask them, hey, have you started to measure yourself against it?
00:19:12.319 --> 00:19:20.799
At least then you can say one they have a framework, so at least they have control points potentially known, not necessarily covered, but known, and they can start to measure themselves.
00:19:20.880 --> 00:19:23.119
Would would that be an acceptable answer to you, Michael?
00:19:23.200 --> 00:19:24.240
Or do you look for more as well?
00:19:25.519 --> 00:19:27.680
I really like those types of controls.
00:19:28.000 --> 00:19:31.599
In addition to that, again, my legal bias is going to come in.
00:19:31.680 --> 00:19:32.079
That's great.
00:19:32.160 --> 00:19:34.319
Yeah, this is all the good contractual control.
00:19:34.480 --> 00:19:38.319
So when I say new, I'm gonna say new non-negotiable.
00:19:38.400 --> 00:19:42.079
So take the time, get with your attorney, and create.
00:19:42.240 --> 00:19:44.960
Here are the things we have to have for anybody who has AI.
00:19:45.119 --> 00:19:46.720
And you can do it like an addendum.
00:19:46.880 --> 00:19:58.240
So I've been I'm I'm experienced enough to remember when bankers went to every vendor and said, Hey, you've got to sign this new GLBA addendum that you're gonna abide by the law.
00:19:58.400 --> 00:19:59.440
Um similar.
00:19:59.599 --> 00:20:04.160
You've got to get them to sign this, and so you can uniformly manage it.
00:20:04.240 --> 00:20:09.759
And if they don't sign it, then you need to kind of figure out another vendor to work with who will sign it.
00:20:09.839 --> 00:20:10.960
So, what are those things?
00:20:11.119 --> 00:20:14.480
You need notice before their material model changes.
00:20:14.799 --> 00:20:20.400
Because if they're m changing their model and it affects your security, you have to know about it.
00:20:20.480 --> 00:20:24.960
And if they don't have to tell you, you are just blindly falling into risk.
00:20:25.839 --> 00:20:28.160
AI-specific instant notifications.
00:20:28.319 --> 00:20:39.839
When their AI does have a problem and they should have a risk model and they should be monitoring it, braccuracy and all sorts of other
corruptions, drift, all the things, you need a very clear path.
00:20:40.000 --> 00:20:46.799
You need clarity on who the subprocessors are, model providers, and audit and evidence rights.
00:20:46.960 --> 00:20:59.039
Auditability becomes king here, especially as you move further down the path of the vendor providing work through their agentic process with a result, which is amazing things, but you need all that.
00:20:59.599 --> 00:21:01.359
Can I can I interject real quick, Michael?
00:21:01.519 --> 00:21:01.599
Sorry.
00:21:02.000 --> 00:21:03.119
So uh I agree with the audit.
00:21:03.279 --> 00:21:06.480
Could I could I suggest two two two different ways to address it?
00:21:06.640 --> 00:21:10.079
One is sometimes the audit is addressed solely for financial purposes.
00:21:10.240 --> 00:21:12.160
I like to use the term security assessment.
00:21:13.119 --> 00:21:17.759
Well, because uh because audit sort of sets people off immediately, they get a little bit nervous and tense.
00:21:17.920 --> 00:21:19.920
Whereas a security assessment isn't an audit.
00:21:20.000 --> 00:21:27.200
I try to always get a separate statement that said was there's audit, which is focused on the financial folks, and then the security assessment.
00:21:27.279 --> 00:21:34.960
And then I can say in the security assessment, if there's been an incident or a potential incident, we can we can we can see you more than once in the year, something like that, right?
00:21:35.119 --> 00:21:36.640
I think that's a very valid point.
00:21:36.799 --> 00:21:38.720
I use that term interchangeably.
00:21:38.799 --> 00:21:41.440
Another phrase I really love is evidence rights.
00:21:41.599 --> 00:21:45.839
I have rights to the evidence of how you validated your AI model.
00:21:46.559 --> 00:21:48.000
I need to see them.
00:21:48.160 --> 00:21:54.720
I need to not just get a letter saying, I did it, because we live in a world where if it wasn't documented, it wasn't done.
00:21:54.880 --> 00:21:59.759
And that's what regulators certainly want, and you all should demand that as well, so that.
00:22:00.160 --> 00:22:06.319
You've got some evidence that this is what I relied upon to understand that this model is valid.
00:22:06.400 --> 00:22:21.680
So I've got that behavioral use case and it creates a framework model that's scalable because now I've got some level of controls coming out of my contract that give me a clear path instead of it being, well, I just hope this vendor works out.
00:22:21.839 --> 00:22:27.519
Now, a common criticism I get from this is, well, Michael, all these things are relatively expensive.
00:22:27.680 --> 00:22:32.960
And a lot of these AI solutions I'm looking at are startups and relatively inexpensive.
00:22:33.039 --> 00:22:34.319
And so that's great.
00:22:34.400 --> 00:22:36.880
You're taking huge amounts of risk.
00:22:37.200 --> 00:22:57.200
And so the reality is if you're a bank, a credit union, a mortgage company, wealth management, you probably don't have the luxury to work with the startup that is doesn't have these things yet because they're too immature in their process, just like you probably wouldn't work with a SaaS startup that doesn't have any security that they can provide you.
00:22:57.440 --> 00:23:04.240
And so you should be really skeptical because you are taking huge amounts of risk if they can't do these things.
00:23:04.480 --> 00:23:07.359
So that doesn't mean you can't ever work with a startup.
00:23:07.519 --> 00:23:18.000
You just need to be really careful so that you don't wind up creating a security breach that costs far more than any savings you're ever going to get out of working with some type of startup.
00:23:18.240 --> 00:23:25.279
If you get into a bind like that, because I've had, you know, in in my practice where the business keeps pushing on you, I don't I don't care what the program says.
00:23:26.160 --> 00:23:27.200
This is the vendor we need.
00:23:27.279 --> 00:23:28.240
We must have this vendor.
00:23:28.400 --> 00:23:34.880
If you're especially if you're not heavily living in an industry like finance or other areas, your program states what you're allowed and not allowed to do.
00:23:35.119 --> 00:23:39.279
It's not that you're saying no, it's that you're telling the business how.
00:23:39.519 --> 00:23:41.680
This is not how you do solve this problem.
00:23:41.759 --> 00:23:53.200
You need to find another vendor to solve this issue because the vendor doesn't meet the requirements that the regulators, and I I know nobody likes to do it to some extent, but the regulators scare the thejeebas out of most leaderships.
00:23:53.279 --> 00:23:56.000
You'll get the attention you need if you use it correctly.
00:23:56.480 --> 00:24:02.480
And even in a world we are living in right now where there's less regulatory pressure on U.S.
00:24:02.720 --> 00:24:06.480
financial institutions right now than there certainly was four years ago.
00:24:06.799 --> 00:24:12.240
But the reality is if you lose all of your clients' data, the regulators are still going to show up.
00:24:12.400 --> 00:24:22.319
So in fact, I actually think it's scarier now because when regulators showed up on a regular basis, you get some forewarning about your process and procedure.
00:24:22.480 --> 00:24:26.319
Now they're just waiting for you to have a breach and then they show up and there's nothing.
00:24:26.640 --> 00:24:27.359
And then the hammer drops.
00:24:28.079 --> 00:24:29.599
The hammer will drop pretty hard on you too.
00:24:31.200 --> 00:24:38.000
One thing I do want to emphasize because it's a practice that is very alive and well in vendor management, needs to stay alive.
00:24:38.160 --> 00:24:40.720
Just because something says AI doesn't mean it's critical.
00:24:40.880 --> 00:24:51.279
For example, if I'm using an AI image generator for my ads, that needs a lot less diligence than an AI underwriting engine to help me make decisions on my loans.
00:24:51.440 --> 00:24:56.720
So that inherent risk, residual risk, all that stuff still applies and can still be super useful.
00:24:56.880 --> 00:25:00.079
So you certainly kind of want to make sure you've got that.
00:25:00.240 --> 00:25:01.920
The regulations will come.
00:25:02.400 --> 00:25:09.440
We're in a low bar right now, but what's making life really complicated, there's state laws out there already.
00:25:09.680 --> 00:25:12.400
Four states have fully enforceable laws right now.
00:25:12.640 --> 00:25:14.480
There are a bunch that are being proposed.
00:25:14.640 --> 00:25:16.559
There's no uniformity at this point.
00:25:16.799 --> 00:25:19.359
The EU has their own set of rules.
00:25:19.599 --> 00:25:21.839
So it's something to continue to watch.
00:25:22.000 --> 00:25:26.720
But I would highly advise you don't wait for somebody to tell you how to do it.
00:25:26.880 --> 00:25:28.799
The risk is real and it's right now.
00:25:28.960 --> 00:25:58.160
So putting good governance procedures in place, figuring out how you're going to handle incidents, what your hard limits are, what contractual controls you put in place, all of which can be extremely helpful so you can govern both data flows as well as your vendors to provide that layer of protection that you're looking for so your organization can run into this technology instead of take all sorts of risks and not really know what those risks are.
00:25:58.400 --> 00:25:58.559
Yeah.
00:25:58.720 --> 00:26:02.079
Well, I I would the analogy I would use the privacy regulations.
00:26:02.160 --> 00:26:04.799
And those are those are a hodgepodge and as well.
00:26:04.880 --> 00:26:06.240
There's not a uniform.
00:26:06.400 --> 00:26:10.079
There's some for industries like GLBA, but you know, the states have a hodgepodge.
00:26:10.160 --> 00:26:13.039
It's not a single uniform privacy across the United States.
00:26:13.119 --> 00:26:16.799
Did people wait for the United States to come up with a single privacy policy?
00:26:16.960 --> 00:26:17.119
No.
00:26:17.279 --> 00:26:21.359
Everybody skated to the California one because it was the most strict for for the most part, right?
00:26:21.519 --> 00:26:24.079
I would suggest you you kind of take the same attitude.
00:26:24.960 --> 00:26:36.640
Find the thing that you think you're going to be the most restricted by based upon your business model and regulatory oversight and skate to that for a while because that's the bet it's your best option, unless you have some other option, Michael, which I'm certainly open to.
00:26:36.960 --> 00:26:42.000
Well, there are lots of AI risk frameworks out there, and it's part of oh, do you like COSO?
00:26:42.480 --> 00:26:44.640
I mean, I won't get into the alphabet soup of them.
00:26:44.880 --> 00:26:46.079
They're all really interesting.
00:26:46.319 --> 00:26:50.240
But pick something and then you can make that your defensible place.
00:26:50.720 --> 00:26:51.039
Yeah.
00:26:51.200 --> 00:26:54.079
But just saying it's too confusing, we're not going to do it.
00:26:54.319 --> 00:26:59.599
Unfortunately, you can't, because most organizations have a vendor.
00:26:59.759 --> 00:27:01.279
I'll just pick on Microsoft.
00:27:01.440 --> 00:27:04.160
Microsoft is threading Copilot into everything.
00:27:04.319 --> 00:27:06.640
Well, how many people use Microsoft tools?
00:27:06.799 --> 00:27:11.119
So your vendors are going to drag you into this AI world whether you want to go or not.
00:27:11.279 --> 00:27:11.519
Yes.
00:27:11.759 --> 00:27:15.039
And it's very expensive to even try and stay out of it.
00:27:15.200 --> 00:27:22.480
And so having the appropriate governance and risk frameworks in place, again, allow your organization to thrive and avoid all of these.
00:27:23.039 --> 00:27:32.079
So, uh, Michael, to help our listeners on the way out, if a risk leader is watching this podcast, what's the one thing they can take back under Nusk and move the needle on GRC and AI risk?
00:27:32.480 --> 00:27:42.000
I think the one thing I would advise is think about what contractual controls you can put in place at the inception of relationships to make your life easier.
00:27:42.240 --> 00:27:49.839
It's a one-and-done kind of thing, instead of, oh, I've got to hire two or three more people, or I'm just getting more data.
00:27:50.079 --> 00:27:53.200
Putting your controls and frameworks in place can help.
00:27:53.279 --> 00:27:54.319
So I guess that's two.
00:27:54.400 --> 00:28:03.119
One's the contract, and the other's the framework, but both are so helpful to helping manage third-party risk in an effective way.
00:28:03.200 --> 00:28:04.880
Uh so I'd highly advise that.
00:28:05.440 --> 00:28:07.359
Uh thanks again for uh tuning in.
00:28:07.440 --> 00:28:09.039
And Michael, thanks again for being a guest.
00:28:09.200 --> 00:28:11.279
It's been really great listening to you and learning.
00:28:12.000 --> 00:28:12.880
My pleasure, Craig.
00:28:12.960 --> 00:28:14.640
Really appreciate you having me on the podcast.
00:28:15.359 --> 00:28:15.759
Thanks.