NVIDIA Corporation (NVDA) Earnings Call Transcript & Summary

Hi, everyone, thanks for joining us today for our webinar, Improve Spear Phishing Detection With AI. Before we begin, we wanted to cover a few housekeeping items. More information can be found in the upper-right corner of the window. [Operator Instructions] A copy of today's slide deck and additional help materials are available in the resource list. We encourage you to download any resources or bookmark any links that you may find useful. [Operator Instructions] If you encounter any other technical issues today, please let us know in the Q&A box, and we'll help you troubleshoot. Now without further ado, we'll turn the event over to our speakers to begin their presentation. Thanks for joining us today. My name is Nicola, and I'm the host of today's webinar, where we'll discuss how generative AI can be used to improve the speed and accuracy of detecting spear phishing e-mails. I'm joined by Adam, who will walk through NVIDIA's AI workflow for spear phishing, a reference example, which can be used as a starting point for building this type of solution. Kris has also joined us to talk about how organizations are applying this technology to build solutions that better defend against spear phishing attacks. As you listen to this session, please feel free to submit your questions in the Q&A window, and we'll be sure to address them, either via the Q&A window or live at the end of the session. The number of cyber attacks continues to increase dramatically every year. Phishing represents the most prevalent and the most expensive cyber threat, costing U.S. organizations $2.7 billion in 2022 alone, and that's up from $2.4 billion in 2021. While generative AI is being used by attackers to create more frequent and more targeted spear phishing attacks, much more can be done to leverage AI to defend against these attacks. Organizations have traditionally relied on employee training to better recognize phishing e-mails or they write rules to filter out suspicious e-mails. With AI, phishing e-mails can be more accurately detected and blocked before they reach a user's inbox. In fact, organizations using AI and automation saved nearly $1.8 million in data breach costs and reduced the time to detect and contain a breach by over 100 days on average. Organizations are recognizing the need to implement AI-based cybersecurity solutions, as evidenced by the fact that cybersecurity is the fastest AI software growth category. In a survey of hundreds of security industry business leaders, the Security Industry Association found that 89% had active AI projects in their research and development pipelines. Real-time monitoring and response to attacks is a primary focus area for cybersecurity AI. Organizations now understand the need to leverage AI to bolster cybersecurity, but spear phishing brings some unique challenges. Phishing e-mails are more generic and designed to scam large numbers of people. And generally, people are aware enough to spot them. However, spear phishing e-mails are customized for specific individuals. They're written for low volume, high click-through. In a spear phishing attack, an e-mail is tailored to a specific persona, job role or industry. For example, a CEO or CIO or someone in the finance department. Because a spear phishing e-mail is so targeted, it tends to be extremely convincing, Often, the only difference between a legitimate e-mail and a spear phishing e-mail is the intent of the user. Spear phishing is, at its core, a data visibility problem. These e-mails are difficult to defend against due to the lack of available training data. Because the attacks are highly personalized, an individual organization will not observe the requisite number of e-mails to train an accurate AI model. The types of spear phishing attacks on financial institutions are very different from health care, for example. Similarly, a spear phishing e-mail targeted at a CFO includes different content than one targeting an engineer. This illustrates the challenge, lack of training data, when addressing spear phishing defense with AI. With NVIDIA NeMo and Morpheus, this lack of available training data to create such spear phishing detectors can be addressed. The NVIDIA Morpheus cybersecurity AI framework provides an NLP model that has been trained using synthetic e-mails generated by NeMo to identify spear phishing attacks. Morpheus is able to infer whether an e-mail is spear phishing or not by first determining the intent of the e-mail. Whether it's seeking financial information, requesting cryptocurrency or asking for personal financial information, such as account or routing numbers. Morpheus then combines the intent of the message with anonymized historical data of the sender, including syntactic analysis, message timing and previous intentions. The spear phishing model incorporates all of this data into an actionable results. Before we talk about our approach to addressing the spear phishing problem with AI, let me introduce some of the technology. Let's start with NVIDIA NeMo. NeMo is the end-to-end cloud-native solution for building, customizing and deploying generative AI models. As enterprises start to take advantage of the transformative power of generative AI, a robust platform is needed to construct and manage these sophisticated models. That's where NVIDIA NeMo comes into play. For the spear phishing detection use case, NeMo's generative AI capabilities can help with creating training data to improve the accuracy and speed of spear phishing e-mail detection. Now I'd like to introduce NVIDIA Morpheus, a cloud-native cybersecurity SDK, which uses AI to identify, capture and react to threats and anomalies that were previously impossible to identify. We believe there's truly nothing out there like Morpheus. The first and most important thing to understand is that NVIDIA doesn't aim to be a cybersecurity company, and we aren't building a cybersecurity solution. Rather, we provide Morpheus to help our partners and customers build better cybersecurity solutions, and here's how we do that. Morpheus is GPU accelerated, enabling the highest performance and massive scale. With Morpheus, all network traffic can be inspected in real time. And with massive data reduction enabled by AI, Morpheus can help to flag anomalies and provide insights so that threats can be detected and addressed much faster. The AI tools integrated with Morpheus make it easier for developers to build cybersecurity solutions that use AI, including RAPIDS, Triton Inference Server and TensorRT. Morpheus simplifies the analysis of logs and telemetry to help detect and mitigate security threats. Morpheus is data-agnostic and can send and receive real-time telemetry from multiple sources, including any NVIDIA GPU and BlueField DPU accelerated server in the data center without impacting performance. Integrating the framework into a third-party cybersecurity offering brings the world's best AI computing to communication networks. With our prebuilt models and AI tools, developers can get a jump start in building applications to leverage AI to address a multitude of cybersecurity use cases, including behavioral AI and sensitive information detection, phishing, fraud detection and ransomware detection. Some of our Morpheus use cases have been assembled into AI workflows. Before Adam talks about our spear phishing detection workflow, I wanted to take a minute to talk about what AI workflows are. NVIDIA AI workflows provide reference examples that illustrate how NVIDIA's AI frameworks can be used to build solutions that address various use cases, including AI chat bots with retrieval augmented generation, speech-enabled intelligent virtual assistance, audio transcription, route optimization, insider threat detection and, of course, spear phishing detection, the topic of today's webinar. And our goal in creating AI workflows is to help developers find a faster and easier path to delivering AI solutions. Using our prepackaged customizable reference applications, which include best-in-class AI software with cloud-native deployable packaging, you can reduce development time and accelerate the deployment of AI solutions. The accuracy and performance of AI solutions has improved with the use of NVIDIA's framework and containers that are performance tuned and tested for NVIDIA GPUs. And when you're ready to transition from pilot to production, you can leverage NVIDIA AI enterprise software, which provides API stability, security and support for all of the AI frameworks and pre-trained models needed for production-grade AI applications. Now I'm going to hand it over to Adam to walk through the spear phishing detection AI workflow.

Thanks, Nicola. To accelerate your path to building this solution, we've developed the spear phishing detection AI workflow. This workflow leverages NVIDIA Morpheus, the open-source, Postfix mail server, a PostgreSQL database as well as a series of pre-trained models using generative AI techniques. Model inference occurs within the Morpheus pipeline, where a score is generated, correlating the classification of the message's intent based upon its content. The end user is then notified to verify the safety of the e-mail, which was flagged by the AI model for intent confirmation. This feedback can then be used to retrain the model and generate additional training e-mails. This workflow is built atop the NVIDIA cloud-native stack with a cloud-native service add-on pack. Cloud-native stack gives enterprises a solid foundation to build and deploy AI solutions on Kubernetes. The add-on pack provides a selection of enterprise-ready services for the workflow, specifically Prometheus and Grafana for monitoring the pipeline's health and throughput, cert manager for providing the necessary security certificates and a database operator used to store statistical information. The core components of this NVIDIA AI workflow are 4 natural language processing pre-trained models to analyze and classify e-mails for identifying spear phishing attempts; the inference pipeline to provide the plumbing from the mail server to inference and back out for delivery; a reference solution for deployment and production, including components like logging and monitoring the workflow, and all of which is available in a cloud-native deployable bundle packaged as a single-helm chart. Using these assets, the NVIDIA AI workflow provides a reference for you to get started and build your own AI solution with minimal preparation and includes enterprise-ready implementation best practices, helping you achieve the desired AI outcome more quickly while still allowing a path for you to deviate. NVIDIA AI workflows are designed as micro services, which means they can be deployed on Kubernetes alone or with other micro services to create production-ready application for seamless scaling in your enterprise environment. The workflow uses many cloud native Kubernetes services, including NVIDIA Morpheus, the Postfix mail server, ML flow, Prometheus, Grafana and PostgreSQL. These components are packaged together into a deployable solution for detecting spear phishing e-mails. The spear phishing workflow components were used to build and deploy the spear phishing inference pipeline shown here. In this reference example pipeline, e-mail is first sent by the Source SMTP mail server and then accepted by an instance of the Postfix mail server. Postfix then sends the mail to the NVIDIA Morpheus pipeline, which is configured as a Postfix content filter. At this stage in the process, the e-mail is enriched with intention analysis and statistical information from the sender's past messages and then inferenced against the pre-trained model to get a spear phishing result. Once processed by Morpheus, the mail is resubmitted to the Postfix mail queue for final delivery at the next [ top mail ] server, for example, Google Mail, on-premise Exchange or Office 365. E-mail intentions play a crucial role in deciding whether the sender's e-mail is benign or spear phishing. To tap into this signal, large language models were trained to infer intents using generated samples of enterprise e-mail content. The NVIDIA spear phishing AI workflow includes 3 intention models, whether the e-mail is acting for a payment, discussing cryptocurrency or requesting personal financial information, such as account and routing numbers. Additionally, the Morpheus pipeline maintains a sketch database to perform historical analysis on a per sender basis. The sketches consider factors like time of day e-mails were sent, grammatical syntax and the intention analysis of previous messages. To give an example, it answers the question if it is common for Nicola, just an example, to send me e-mails at 2:00 a.m. on a Saturday with poor grammar asking about Bitcoin. It isn't. These become features of the spear phishing detection pre-trained model, which is included in this phishing detection AI workflow. Note that the pre-trained models provided were trained using generative AI in such a way that no personal identifiable information or sensitive information are included. The spear phishing detection AI workflow can be easily integrated into your e-mail solution. However, we've also included a demo e-mail environment for convenience. Here's an example, e-mail identified as spear phishing, and you can see the modified subject line indicating the result. The pipeline has also included some additional information in the form of status headers. The status headers indicate the inference result from the overall model, in this case, a 99% confidence, the configured minimum threshold to identify spear phishing e-mails and the result of each intention classification. In this case, the e-mail was flagged as both banking information and cryptocurrency. A 0 indicates a positive result. Rules can be easily configured in your mail system to automatically quarantine or block mail identified as spear phishing. Using generative AI and the NVIDIA Morpheus cybersecurity AI framework, developers can build solutions that detect spear phishing attempts more effectively and with extremely short training times. In fact, in our testing, we were able to detect 90% of targeted spear phishing e-mails after less than 24 hours of training, a 20% improvement compared to a typical phishing detection solution. Morpheus also enables profiling of interactions between individual users, adding another layer of protection in detecting anomalies and threats. You can apply to try the spear phishing detection AI workflow in our hosted environment, NVIDIA LaunchPad. Now I'd like to introduce Kris, who will talk to you about how he's working with organizations to implement faster, more accurate spear phishing detection with NVIDIA AI.

Thank you, Adam, and thank you, Nicola, and hello, everyone. My name is Kristopher Keith, and I'm on the NVIDIA public sector team. And I'm excited to be speaking with you today about how generative AI is unlocking new ways to defend against an evolving threat, and that is spear phishing. I also want to thank our federal customers and partners for dialing into the webinar. I encourage you to ask questions in the Q&A. This is my personal invitation to join me at GTC in March. It's our annual conference. It's going to be held in San Jose, California again. And we're back to being in-person, so I'm expecting it to be a sold-out conference this year. And there will be plenty of cybersecurity content and guest speakers. Hope to see you there. So let's take what Nicola and Adam have already shared and bring that into the public sector. For those of you who are new to working with NVIDIA and cybersecurity in the public sector, let me first share that we view cybersecurity as a data problem. In spear phishing examples, there simply isn't enough data to train effective AI models, so we leverage the power of generative AI to fill that gap. Now I've spent a lot of time recently talking to federal customers who are implementing zero trust solutions, and this capability aligns perfectly with the pillars around AI and ML requirements for automating things like threat detection. And the topic of leveraging generative AI has come up a lot as well lately. The rise in spear phishing attacks against government agencies and critical infrastructure, we're seeing demands for new technologies like NVIDIA's Morpheus. I was at a Zero Trust Cybersecurity Summit in Washington recently. And one of the agency CISOs was on the panel, and he said it pretty plainly, "If AI is being used to create attacks, we have to use AI to defend." Now if we zoom out across all of the federal government, let's talk about how big of a vulnerability this really is. We're dealing with sophisticated attacks, convincing e-mails that can be created faster and easier, and just about any level of adversary can deploy this. What's more, we're talking millions of potential victims. Think of all the .mil or the .gov, the supporting supply chain of contractors and vendors. This is a complicated problem and conventional defenses of training and awareness, looking for poorly created e-mails or misspellings or software with rules is no longer sufficient. We need to stop malicious e-mails from getting into inboxes so our employees aren't left making the decision if this e-mail is malicious or not. So I'm sure by now, you're probably already thinking about your own use case. Here are just a few that we've been looking into as well. You may have seen alerts from CISA and IRS. CISA is the Cybersecurity and Information Security Agency (sic) [Cybersecurity and Infrastructure Security Agency]. They publish warnings to tax professionals and IRS employees to be extra vigilant during tax season and don't click on any links and e-mails. For how many of us at this time of year are getting e-mail saying, "Click here to download your W-2," but are we all trained not to click on links? Is this a well-written e-mail from HR or an AI-generated spear phishing e-mail? It's just getting harder and harder to tell. Think about someone stealing credentials of a tax professional or us, as employees, and having that information in the hands of adversaries with all of that PII data. Another example, what about the end of the government fiscal year? When everyone is under pressure to process last-minute activities, you've got government contracting officers, contracts administrators, vendors, subcontractors, accounts payable, receivable. There's a lot of opportunity to fall victim to an adversary, who wants to do something like change payment routing information for a government contract. Yes, these are real-life examples I'm talking about. Key here is to basically reduce the number of phishing e-mails that ever get into employee inboxes. Once someone clicks on a link, that just might be too late. They compromise their credentials, open themselves to a ransomware attack or worse. So I encourage every organization in the public sector supply chain to be thinking about how to deploy the best defenses out there to keep their data secure and protect against financial loss against this evolving adversary capability. So with that, I'm really looking forward to working with our customers and partners to do amazing work to defend our most critical data, and I hope to see you at GTC. We're going to take some time now to answer questions you've submitted during our session. But before we do that, I'd like to highlight some resources that you'll find helpful as you embark on your cybersecurity AI journey. The week of March 18, NVIDIA's GTC AI conference is going to be held. It's in-person at the San Jose Convention Center in California. We've planned over 600 sessions, 200-plus exhibits, networking events around AI and accelerated computing. For more information, visit nvidia.com/gtc. That's nvidia.com/gtc. I hope you'll join us. If you'd like to get more information about our AI-based cybersecurity, visit our Web page at nvidia.com/morpheus. And lastly, if you're ready to get started with spear phishing detection AI workflow that Adam reviewed today and you want to try it hands on, you can visit our LaunchPad site at nvidia.com/try-morpheus. Thank you. Speaking to GTC, we have a ton of great content planned around cybersecurity and generative AI. Sign up to hear our Head of Cybersecurity Engineering, Bartley Richardson, talk about how we're helping organizations to apply generative AI to improve their cybersecurity posture. And if you like what you saw today with Adam's walk-through of our spear phishing detection AI workflow, sign up for the hands-on lab experience and bring your own laptop so our solutions architects, [ Adio and JB ], they can walk you through how to build an application for AI-based spear phishing detection on your own. Hear from 2 machine learning developers, Colin and David, who are in the U.S. Navy and how they tackle the issue of deploying long-context LLMs. Stop by our connect with the expert sessions, where cybersecurity and generative AI and threat hunting experts are going to be on standby to answer all of your burning questions. And lastly, register to attend our Cybersecurity Developer Day. You'll learn about how we're creating workflows that apply generative AI and LLMs to address different cybersecurity challenges.

So thank you, everyone, for joining us today. I'm seeing a lot of questions that are starting to populate into our chat window. So thanks so much for the questions. We want this to be interactive. So continue to post your questions into the chat. And as you can see, we have several people here to help with answering your questions today. I'm Nicola Sessions, as mentioned earlier in this webcast. I'm joined by Kristopher Keith, Adam Wood, and Chau Dang, would you like to introduce yourself as well?

Hi, everyone. I'm the product manager for Morpheus. And our latest AI workflow is spear phishing, and I'm excited to answer any of your questions here. I'd like that we have a good crowd.

So before we get into the questions that you guys have rolling in, so thank you so much for all the questions, we wanted to start with a couple of poll questions and give you an opportunity to give us a little bit more insight into where you are in your cybersecurity AI journey. So the first question here is, which of the following describe your organization's experience with cybersecurity AI? So if you guys want to just pop your answers and select the ones that apply to you here, that would be great. We'll give you a few seconds to answer those. [Voting]

And I just wanted to say thanks for all the good, really good questions and your time today. We really are encouraged by some of the experts that we've gone on, both from public sector, private sector. We see folks from financial industry, health. Obviously, spear phishing is a really big problem across every industry. And in my role in supporting the public sector, I've really seen some interesting use cases. So just really looking forward to engaging with you guys. I hope you can -- as you saw, we're really looking for some hands-on training at our GTC Conference. So thanks, everybody, for joining in. I hope you can -- we can get to all these questions. They're coming in, in droves here. So we've got Adam on standby to help us with some of the technical. He's one of our solutions architects on this.

Hi, everybody.

Yes. So I think what we're waiting for the next poll question, maybe I'll jump to one of the questions that has been coming in on the chat just so that we can start to get through some of the stuff that -- your burning questions and things that you're asking. And the first one is, thanks for the overview of the spear phishing Morpheus analytics. Is the latest Morpheus release available, 23.11? And so that's great. Yes, 23.11 is our latest release, and it is available. So you can get it on NGC. You can get it on GitHub.

Right now, I would recommend you use it on LaunchPad. We're still working on some of the updates for NGC catalog.

Yes. And so to Chau's point, yes, that the best start play -- starting point for spear phishing workflow is to just go to the LaunchPad lab, and then you can get kind of like a guided walk through there in a demo environment, in our own hosted environment. And then, of course, from there, you can try it in your own infrastructure. So [ Cath ], are we ready to go to the next poll question here? And looking at the first poll question, it looks like a good deal of you are already evaluating cybersecurity AI and looking for best practices. So that's a promising and exciting start. So the next poll question, which of the following describe your organization's experience with NVIDIA's cybersecurity AI framework? So while you guys are working through your answers on that next question, we'll take a look at some of the other Q&A that's come in during the course of this webinar.

Let me ask Adam one that I get often, and I think we got asked it in the chat is, do we have to have different intent models for public sector versus private sector? What are your thoughts, Adam, on the kinds of intent we see across different -- kind of what do we say accounts or markets or sectors?

Yes. That's a great question, Kris. I think it's really going to depend a lot on the types of threats that your organization sees. So in our case, we focused on banking, crypto, financial details because we see a lot of that in the spear phishing e-mails that we're getting. If you're in a different industry or you see different types of threats coming in, the solution is very flexible. So you can train your own intent and infer against that, right, and that can be part of your spear phishing pipeline?

Yes, I know that in public sector, some of the use cases are around critical infrastructure, [ SCADA ] systems, kind of government-related domains. And so, yes, with the ability to tailor like Adam mentioned, you really -- it's up to whatever organization you want to be able to kind of customize or train some of these. That's really the power of generative AI. What about accuracy? Or how are we measuring the synthetic training data results versus real spear phishing attacks? Do we have any numbers to show or anything like that? Because that kind of -- that question comes up a lot, and it's come up in the chat a few times.

Yes. When we compare it to just our -- the generic phishing solutions, we see that we're able to actually capture 90% of the spear phishing e-mails. And so that's like about a 20% improvement. And so for us, it's really good. In addition to that, we actually had a feedback loop where we work with the [ Red team ] and we're able to then filter out those spear phishing e-mails as well, and we can get close to 100%. So it's quite powerful.

So while you guys are working your way through the questions here, I'm going to disclose the results of the latest -- the second poll question about which of the following describes your organization's experience with NVIDIA's cybersecurity AI framework? And it looks like it's about 50% have heard of it, but are still looking to try it. And the great news is that we have multiple ways for people to try our NVIDIA Morpheus framework. As we mentioned before, LaunchPad is a great place to get started in a hosted environment. And then obviously, we also have GitHub, and we have containers and NGC. And then there are a great deal of you, I think, on this call who are learning about this for the first time. So keep your questions pouring in, and we will continue to answer them. And I see the -- one of the questions I just answered, is it available in NGC? Yes, absolutely, the workflows for spear phishing detection. We also have a workflow for insider threat detection that we call digital fingerprinting. That one is also available in LaunchPad, in NGC and in GitHub. So there are plenty of places to get started and start using these. Are there any other questions in the list here that you want to take and address next?

I see one, Nicola. The question is, how is the improved models? How are they pushed to clients? So we use ML flow to handle deploying and updating models within the pipeline. And so as you add a new intent or something like that and you've trained the spear phishing model, you can republish that through ML flow, and the pipeline will pick up the latest version automatically.

Yes. One of the questions that I see that I would just want to drive the point home is that asking about techniques as to why this is an enhancement. And as mentioned by Kris and Adam earlier, it's around the intent. But whatever the intent is, you can actually generate additional synthetic e-mails around. And so the idea is that what you have in LaunchPad right now is just the crypto and banking, but you can add additional intents. So if you can imagine, I don't know, if there's a war or any kind of like current events that are happening that trigger new phishing e-mails that you can now create a new intent to be added to your system and then improve the accuracy.

And there's a couple of new questions that have come in. Are there any use cases published using Morpheus? Yes, on our nvidia.com/morpheus website, we've got a couple of use cases that we've published there. So you can see some examples of the realm of the possible with Morpheus and ways to get started and some inspiration. We have sensitive information detection. Chau mentioned earlier, crypto mining, malware detection. We talked about spear phishing, ransomware detection. And that ransomware use case actually ties into another question that has come in here about the role of DPUs in some of these use cases. And so -- I mean, DPUs are not required for Morpheus or for most of the Morpheus use cases, but in specific use cases, DPUs play a very valuable role, an important role. So Chau?

So the additional use cases we have is just like if there is -- if within your company, there's a policy around crypto mining, right, whether it's prohibited or not. And so we're able to take in data from the DPU and process it and determine whether or not there is any crypto mining based on the memory usage. And so the idea there is that you can ingest any data from the DPU as a data source. So any sort -- if you are thinking of any uses where you are using DPU as a data source, you can then use Morpheus as a pipeline for processing or doing any kind of analytics on that data set, and that's how we've used it here. And then there was another question about customers who have tried this and partners who have tried this in their environments. And I think a great place to start for that information is NVIDIA GTC. So I think people know that we have a GTC coming up in a couple of months, in March. And we have some exciting content planned for the upcoming GTC, where we will have partners and customers showcasing how they've been using our technology and the different use cases that they've applied this to. So that's a great place to get started. And also, you don't have to wait until the next GTC. If you look at our GTC on-demand archives, there are also additional examples of customers who have given talks and talked about their experience with Morpheus, retail customers for general phishing detection use cases, financial services customers for ransomware use cases are some examples. And so yes, if you want to take a look at that content, I think you'll find some interesting use cases and examples there. And then, of course, we hope that we see you all at GTC in March in some way, shape or form, whether it's virtual or in person. I see a question here about our models and whether or not they are just from Hugging Face, an OpenAI or are they our own? Some of them are our own, right? And so we will identify that, and we have model cards for them to give you all the details on the model. But we take usually the public data sets that we train our models on.

And then there's another question about which AI technologies are we talking about here. And so here, in this webinar, we're talking about a use case or an example or workflow that we've built where you can apply the Morpheus-accelerated cybersecurity AI pipeline combined with NVIDIA's NeMo generative AI framework to address the issue of spear phishing detection in your environment. And so the idea is that you can pull in multiple different pieces of NVIDIA technology to help build solutions that will help address issues like this. And so we provided an example. We've got a great deep partner ecosystem of -- we've got a great ecosystem of partners that can help with implementing these types of solutions, and we provide all of these tools on NGC on GitHub. So if people want to get started and take a look at these things themselves, you can find documentation, you can find notebooks, you can find all kinds of tools to get you started. And of course, like we've mentioned on the call a couple of times, our LaunchPad Lab. So kind of a hands-on demo hosted environment, so people can kind of get started and see. Yes.

Next, there's a question [indiscernible] interesting, how accurate is synthetic training data versus real data? The question is interesting is because no real data on spear phishing, right? And I think that, that's why generative AI is used here is because you don't have the amount of data required to train a model for a lot of spear phishing. And so that's why you need a synthetic data. So there's actually no comparison in that case.

Yes, I get that a lot. As I said earlier, I work with a lot of public sector customers, and there's just a severe lack of training data, or every customer I work with says,my e-mails are unique, my staff, my work, my mission is unique, and so having something where they can train their own model, where they can kind of tailor it to their mission. I work across 1 billion agencies. That's a very different issue. There's a lot of privacy issues. There's a lot of things in defense, it's very different. So whether it's -- whatever the kind of -- from a public sector domain, we have very unique -- we have a unique mission sets and unique organizations within those companies, whether it's the finance department, the C-suite level leadership, the flag officers. A model can be tailored towards any of those populations. That really generative AI without generative AI, we really couldn't do that before. So I'm finding that that's the most -- that thing resonates the most with them. It's spear phishing is one of those ubiquitous use cases that everybody is potentially victim to. So we certainly started off with that. But I don't know where we're going to be in the road map, but generative AI is going to unlock a lot of use cases like this. I see another question, are the models compared against each other? And the models themselves are not compared against each other, but the results of previous inferences are. So on a per center basis, we keep track of the types of e-mails that they're sending, and so we can compare like is it normal for that person to send e-mails about banking, right? Because obviously, people do work in banking. And so it might be normal for them to have conversations about those types of things, and that wouldn't necessarily be anomalous. And so we keep track of things like that, and grammatical syntax as well as when they typically send e-mails. And so that helps give -- it lends itself to the accuracy.

So one of the questions is, is Morpheus a private solution or could I implement other solutions? And so Morpheus is available on GitHub and NGC, and so it's openly available. And I think one of the values is that it's meant to interoperate or integrate with existing technology. So I don't know, Chau, you have data points to add around that?

Yes. So the thing -- we can. Actually, Adam, I think, could you mention like the integration points in the stages. Could you speak on that?

Yes. The pipeline is written in Python, it's built atop the standard Morpheus SDK, right? So we can integrate any number of existing stages or create new ones relatively easily. It's all open source, as we mentioned before. So it's on GitHub, and we can -- it's easy to look at the other opportunities for integration. But for example, in the digital fingerprinting workflow, we integrated with Elastic Search, right? So theoretically, you could have e-mails or the results of the spear phishing send over there. Or if you have another solution that would be easy enough to make those same changes and have the alerts feed into that system. Morpheus is really flexible that way.