Coursera

Fortune Editors

Fortune's annual list dedicated to the idea that business is uniquely suited to address society’s biggest challenges.

CTW.1122.Coursera

Coursera works with employers and higher-education institutions to offer online courses, certificates, and degrees—with an emphasis on providing people from underserved communities with a path to better-paying jobs. Over 2,000 courses are free, and most can be audited; its enterprise partners often pay the cost of classes. Enrollment soared during COVID, with all-time users reaching 107 million in June.

Change the World

Change The World

Impact Segment

Sector

Coursera (COUR): Company Profile, Stock Price, News, Rankings

legacy

coursera

Coursera (COUR): Company Profile, Stock Price, News, Rankings | Fortune

Why two-time founder and MacArthur Genius Daphne Koller sees a future where humans ‘create together with the machines’

/2025/07/09/leadership-next-daphne-koller-insitro-coursera/

Preston Fore

This CEO went from folding clothes at Nordstrom to being Jeff Bezos’ right hand man—the billionaire told him to delegate to his employees more

/2025/07/06/coursera-ceo-greg-hart-jeff-bezos-technical-advisor-effective-leadership-delegate-decisions/

OpenAI partners with Wharton for a new course focused on leveraging ChatGPT for teachers

/education/articles/openai-chatgpt-university-of-pennsylvania-wharton-course/

Jeff Maggioncalda

Coursera CEO: AI won’t replace teachers—but it will transform education

/2024/09/24/coursera-ceo-ai-teachers-transform-education-tech/

Higher ed leaders can learn generative AI for free with this new course

/education/articles/exclusive-generative-ai-university-leaders-course/

https://fortune.com/education/articles/openai-chatgpt-university-of-pennsylvania-wharton-course/?itm_source=parsely-api

This CEO went from folding clothes at Nordstrom to being Jeff Bezos' right hand man—the billionaire told him to delegate to his employees more

https://fortune.com/2025/07/06/coursera-ceo-greg-hart-jeff-bezos-technical-advisor-effective-leadership-delegate-decisions/?itm_source=parsely-api

https://fortune.com/2024/09/24/coursera-ceo-ai-teachers-transform-education-tech/?itm_source=parsely-api

Why two-time founder and MacArthur Genius Daphne Koller sees a future where humans 'create together with the machines'

https://fortune.com/2025/07/09/leadership-next-daphne-koller-insitro-coursera/?itm_source=parsely-api

Daphne Koller, CEO of insitro and cofounder of Coursera, joins Leadership Next to unpack a bold idea: Just as calculus revolutionized our understanding of physics, AI could do the same for biology. In this episode, Koller explains how insitro is blending biology and machine learning to unlock new treatments for diseases like ALS and fatty liver. She also opens up about her leap from academia to entrepreneurship, the ethical challenges of data collection, and why the future of medicine lies in personalized, AI-driven care. Plus, her take on why humans and machines must work in tandem to drive meaningful innovation. 

The Machine Learns Medicine: Daphne Koller’s Vision for AI in Biotech

Photo Courtesy of Daphne Koller; Graphics by Fortune

Leadership

How insitro’s founder is using AI to decode disease—and why she sees it as biology’s version of calculus. 

Pat Geraghty, President/CEO Guidewell, Jeff Tarr, CEO, Skillsoft, Terry Rasmussen, President/CEO Thrivent, Glenn Fogel CEO and Director, Booking Holdings, Roberto Martinez, President, PepsiCo Mexico Foods

The Exponential Power Of Generative AI In Business Part 2

Uncategorized

How AI's Reshaping the Business Landscape

Pat Geraghty, CEO Guidewell, Jeff Tarr, Skillsoft, Terry Rasmussen, Glenn Fogel, Booking Holdings, Roberto Martinez, PepsiCo Mexico Foods

How are you thinking about incorporating this into your programs? Well, we're focused on corporate learning learning inside the enterprise a little different than Shag's model and we've been using A I for a long time. Generative A I a less long time, but we've been at it since before. It made the news. We believe that it is going to impact almost every knowledge worker job to some extent. And so that creates a tremendous imperative for reskilling and up skilling. And so it's changing what we teach and it's also changing how we teach. And when I say what we teach, when you look at learning inside the enterprise, most of it falls in the three buckets, leadership and business skills, technology training and compliance. And all three categories are being impacted because learners need to understand it and their employers need, need their employees to understand it. Our introduction to a generative Generative A I course had was four times at four times. The enrollment of the next most popular course of all time for us. Our new course and new learning journeys on prompt engineering, advanced prompt engineering, uh ethical A I generative A I for decision makers and leaders are are of great interest and generating a tremendous amount of, of usage. Um And in fact, just about every course we developed going forward, we're, we're actually asking before we begin production on a course is generative A I uh relevant to this subject area and making sure we hit pause before we move forward so that we're not missing an important new technology. As for how we teach, it's already being used to accelerate localization and translation because it's not just about the words, but it's also about giving it cultural context. And now we can do that for the first time using machine learning, using this gene A I technology. Uh We've been be testing a generative A I coach that is walking learners through actual virtual simulations. Now, a lot of the work we're doing is on guard rails and filters to make sure that we're looking out for bias and inaccuracies. But we can do that in our controlled environment. And we're uh we're about to move into a beta test with more than 50 large enterprise customers. And uh and then also I would say finally, uh we see the technology is having great promise for what's been the holy grail of online learning and learning in general, which is personalized learning at scale that adapts both the both the content and the assessments adapt to the needs and learning styles of the individual and we're not there yet. But this is turbo charging work in that direction and we're really excited about what we're seeing. Yeah, fascinating. Uh Terry Rasmussen, maybe you can uh comment on, on what's happening in the financial world. Happy to um let's see here. Hold on. So in, in financial services, I do believe that generative A I is a powerful tool and it has the potential to, to really transform, I think every business um in, you know, out there in financial services, I think the best use case for thriving is with our financial advisors and it can be used to, it's that co-pilot concept where it can be used to, to, to free their time and so that they can spend more time serving clients and giving great experiences. That's right. And we kind of think of um our advisors as being experts in head math and heart math and in head math, I think generative A I has just a huge potential to have offer that personalized plan if you will. But then it's our advisors that come with the heart mass and, and I think it will be years before generative A I can replace the heart mass because the heart mask doesn't always align with head mask. And so in heart mass, it's really what's the most important uh for this individual or this client that we're serving. And so I see it as, as again, an opportunity to really come alongside our advisors and, and as a copilot. Yeah. And Terry, how are you thinking about the issue of transparency when you're, when you're using the technology? Are you quick to tell the, that your customers that you're using it? Is that an important part of the trust equation? Oh, absolutely. You know, we, we own trust in the marketplace and financial services and so transparency is really critical for us and, and, you know, we, we just recently rolled out guidelines for the organization um similar to, to make sure that, you know, the use um that our use is appropriate. And so we've got kind of four guidelines at a high level. Uh first you went and out, leave it out. And so that really is the proprietary information, making sure that, you know, we're, we're sensitive to that and we are retraining. I agree. I think it was Chris who said that, you know, this, you should always protect your proprietary information, but it's, it's good to remind people of that. Um We also believe trust is earned and not implied. And so again, you owning trust in the marketplace, it's critically important that, that we earn that trust and continue to earn that trust. Um and then scan and review before you use. Uh that is something that we're, we're telling our workforce, make sure that you're understanding the use. And then always remember our trusted reputation. It's the most important asset we have in the marketplace is that trusted reputation. Yeah, thank you uh uh Glen Vogel II. I know you've been using A I for a long time in your bookings business, but generative A I is a different animal. Uh uh where, where, where do you see that playing into your business? Well, what it is doing or what it will do is get us much closer to what our vision is of what I call the connected trip. We all know frustrating is to plan your travels. And we all remember the old style old school travel agent who is a human being, knew something about you knew kind of what you could afford. We do it all for you and everybody when you are putting together your travel now using a digital method, it is frustrating. It takes forever. I wish somebody else would do it for me. Well, that's what Geneva is basically going to do. You're gonna have that conversation again. Always. It's gonna be with an A I agent and it's going to personalize as was pointed out earlier. How important is the personalization really being the focus on an individual what they may like? And I can give a little example of every single um search result shows static content in travel. That is the same for everybody. I have to like going to a gym. I want a hotel as a gym. My wife likes high tea in London. Imagine that when it is going through the process of what we're trying to discover it is customizing the content describing which hotels have these things for my wife and I versus somebody else who's interested in something else. We'll see different content that's really going to be something fantastic. And then of course, men, customer service, we all know the frustration of customer service. It's very difficult generative A I is going to improve this hugely, being able to know what the problems are possibly seeing what they may be before they happen and being able to respond much faster than we currently are able to do, which would be wonderful. And how quickly are you implementing Glenn? So we have two check T BT plugins. We have kayak an open table are already, that's how that's in the wild right now. Um We are, you know, moving methodically and carefully want to do everything in a responsible way and of course, that trust issue in the financial area, certainly true. But it is also important in travel. Nobody wants their travels screwed up. So we have to be very careful in that to make sure that we don't ruin the trust that we've built over many, many, many years. Um Roberto Martinez, can I bring you into this conversation? Would love to. Yeah. Yes, of course. Good morning. Everybody would, would love to hear how it's uh uh there's been a lot of talk about the application of this technology and marketing, would love to hear about that or other applications that you think are promising. Yeah, in our case, uh I'm, I work for Pepsico in Mexico so that everybody knows who I am. Um in our case and, and what Joe said at the beginning of the conversation right now is more around. How do you make sure that we are uh secure in terms of the information that we use there and, and just be careful with the teams on, on what they put out there. A couple of things. You know, we, we think this obviously will, will help us be more efficient, reduce costs and obviously be closer to our consumers and our customers. Um Some of the things that we're looking at is, you know, um product development and new flavors, a new propositions to the consumer, but also using it in our case, um uh specifically for more things around the sales cycle. Um in terms of, you know, we have a lot of mom and pop stores that we serve in Mexico, about 800,000 stores we serve uh here. So how do we, you know, make sure uh that we can have the right assortment, uh the right velocities, what is selling here, what is not selling here and then build that into an E I engine so that we can um a little bit, you know, have a, a better, better service to the customer, better service to our, to our sale sales rep. Um The other part is I would say is, um you know, as I, as I said before, connecting to the consumer, that's a little bit, you know, just starting. So not, not out of there. And then, you know, as somebody said before, we wanna make sure that we're responsible, how we use this and our teams are responsible and how we're using this. So a little bit of latitude but within a framework, uh just because we've, we've also seen some examples of other companies just, you know, don't have. Uh and we're all starting to build, let's say internal policies if you want, but just making sure that our teams uh do not put, you know, um you know, uh confidential information out there because then it's, it's already out there. So that's a little bit of the, the conversation right now in inside of principle.

The Exponential Power of Generative AI in Business Part 1

The Exponential Power of Generative AI in Business

Pat Geraghty,CEO Guidewell, Jeff Tarr, CEO, Skillsoft, Terry Rasmussen, Thrivent, Glenn Fogel, Booking Holdings, Roberto Martinez,  PepsiCo Mexico Foods

Chat GP T has enjoyed this sort of uh stratospherically fast adoption in the consumer world, fastest adoption of any new technology in history. I can't keep up with the numbers myself, but it's been kind of instantly omni uh present but big business as far as I can tell from various conversations I've had and from some new polling data we have is still being kind of cautious, not about A I generally uh predictive A I data analytics. That's all moving ahead but about generative A I and how to use generative A I in business. I assume you're having a similar experience to what I'm having. And if so, I wonder if you could explain why the caution. Yeah, we are allen. I think it's a great, it's a great question, a great observation as to what's happening. Um There's no question, generative A I and, and particularly Chat GP T is having a moment and that, that moment's been unfolding over the last few months. And I think the reason big businesses are cautious is the same reason that it's having a moment which is, um, largely people were awestruck about the capability that they were seeing and the accessibility of A I tools in a very, very different way than what has been the discussion over the last few years. And those in the tech community know that that didn't happen overnight. And this investment in this capability was being built over time. But for many people who were not close to this technology, suddenly you had this, this moment that unfolded as people watched the capability of generative A I to deliver really, really thoughtful responses, sometimes fully accurate, sometimes not fully accurate. And it, and it opened up this question about, well, how do we use this? Clearly, there's power here. How do we use this the right way? How do we use it in a responsible way? And every single client I'm talking to is trying to figure out how to go fast but how to go fast in a responsible way right now. And so Joe, if you had to sort of pinpoint the three or four things that are causing that caution, say, yeah, we gotta make sure we work out ABC D, what would you, what would you say? Well, I think one is clearly um confidentiality in the proprietary information. How does that actually work in an environment that is by its nature has, has elements that are open. And so everybody is asking the question if I'm really going to use it for my business, I wanna put business data in there. But if I put business data in there and it's in the public domain that, that doesn't work out for any of us. So that proprietary data question is, is clearly top of mind. The second one is, what am I using it for? And I'll, I'll, I'll describe it as the right use for the right business case, the right use cases. And I think because it, it feels very new to a lot of people. They're not really sure what the applicable use is, they see the power, but they're not sure how to direct that power to the right things. And that, that leads me to the third piece, which is how do I prepare my people for what's happening? Um You see all kinds of estimates of what the implications are going to be on productivity, on the ability to take tasks off of people's plates. Well, that converts into people's jobs and that causes large organizations to take a breath and figure out how that all plays out. So preparing people for the change and in our perspective, and many of you probably saw the PWC announcement. We, we think that comes down to training people to take advantage of the technologies because we know that if you don't prepare people for the technology, it's coming anyway. And so part of the responsibility we see for any organization, but certainly big employers is to help your people prepare for the changes that are coming. Yeah, that's good. Thanks. Uh Chris. Let me uh let me turn to you. I wanna go into each of those uh things that Joe mentioned. But I, but I, but I'd love for you to start to just, you have these conversations every day. I would imagine. In fact, I doubt you're getting much sleep these days. Uh But uh I, I wonder if you could just talk about why this moment is so important to business. I think we see the transformative potential of A I and, and it's, it's not only transforming the tech landscape, but it's gonna transform business, it's gonna transform much of what we do as humans. And, and this, as you mentioned, differentiating what we've known as A I, which is more around machine learning and predictive technologies from what we've seen to where we are today, which is really the generative A I era, which is truly changing many, many things. And, and you know, what we see is really a, a bifurcation. As you, you know, Joe, I think well stated some of the, the ways in which larger organizations are approaching A I, you know, certainly in the tech landscape and, and, and the start up world. Um you know, we've seen literally thousands of start ups in the last six months move into this space. Uh There's just a tremendous amount of speed and innovation happening. You know, I happen to sit here in San Francisco and there's literally a hackathon happening nearly three or four times a week and it, it's just the, the amount of energy going into the space is palpable. Uh, and it's, it's, you know, we, we see the innovation opportunities. You know, there's, there's, you know, companies going after marketing, there's companies going after customer support opportunities. Uh There's a lot of tooling, focus areas, uh, speech recognition, you name it and it's just, there's, there's just a number of different approaches that, that organizations are taking and, and I wouldn't say that all large companies are, are being cautious, you know, for example, um you know, we've already been working with companies like Carmax as an example that are, that are using generative A I as part of their uh review process and making, you know, really being able to synthesize the information better for their buyers. Uh Epic is another one in the health care software space that's uh that's been working with us on, on A I and certainly PWC and Joe and his team uh are working with us in this, in this like is in this regard as well. So we, we are seeing movement here uh in companies of all sizes. Can you talk a little bit about the first issue Joe raised which is protecting proprietary data, the confidentiality. Uh uh there was a story this morning that says Apple has told its people to not use chat GP T. Now that may be because they have an alternative technology that they want to adopt. But in general, I find a lot of business leaders worried about mingling their proprietary information with the public information that's in these giant large language models. Proprietary information is certainly got a lot of value. And we, that's a common conversation to be having right now with any organization is. How do I make sure that I have good protections, good safeguards around my information. I will just state for a moment by the way, that, that, that was true before. Generative A I, that was, there was a lot of concern in most large organizations around uh protecting their data. But I think that's more, you know, I think most or organizations are trying to understand in the era of generative A I and if I'm gonna use generative A I, how do I make sure, how do I ensure that my information is well protected, it's utilized effectively. And so as I think Joe said it, well, most organizations are just trying to understand exactly how that's gonna work. And that's something, you know, we've, you know, we spend a lot of energy on that with our customers and helping them understand and there's different methodologies that you can use. You'll hear, you'll hear terminology like fine tuning the model. Uh you'll hear terminology like grounding the model. Um There are these are different ways in which organizations can use um generative A I models, large language models uh to, you know, on their own information without passing that information back to the, to the base level model itself. But it's a solvable problem. In other words, it's absolutely a solvable problem. And, and then the other, the second, uh, issue that Joe raised was the, in a way the enormity uh, of potential for these models. I think people have a hard time figuring out. Where do I start? Where's the highest value? What's the, what's the first place where I should apply this? And how, how do you, uh again, I'm sure you're having these conversations every day. How do you walk companies through that uh conversation? They're, they're absolutely, there's absolutely a progression to this, that, that organizations I can think about. I think at the, at the, at the base level, if you will, what I always tell people, certainly leaders is the first thing you can do is just familiarize yourself with using it. That's one of the great things about having, you know, bing chat or chat GP T and some of these tools available in the market. If anyone can use it, a leader can use it. You know, Alan, you could use it to develop the questions you might ask us today. Perhaps you've done that. I did. Yeah, these are not my questions, but you're doing great. So keep going. Uh And, and that, you know, that's a, that's a, just a great way to familiarize yourself with some of the ways in which um these products can work. And then II I think from a, from a productivity perspective, you know, we've approached this by saying the first step is really these co pilots. You know, the first product we brought to market was something called github co pilot. It's a tool to help developers with their co you know, writing code. It, it works alongside the developer, uh the engineer and we're bringing that co pilot capability to a variety of different applications, whether it's Outlook or Word or Excel uh powerpoint. Uh You can obviously we, we're even bringing it to, to cyber security. That was a topic you brought up earlier to help uh those who work in your security operations center to be more effective at their role. So the first, the first step in our in our approach is product activity, bring a co pilot along with you. It'll make your knowledge workers more effective, help them get up to speed quicker on topics. And it's like it's like having this really, really smart assistant who can help you do many things uh in a very short period of time without you having to go out and research large amounts of information. And then from there, you can see how um ultimately you can use this in customer facing operations, you know, customer support category that we believe is gonna have a lot of application for generative A I I think it's an there's opportunities there to provide even better experiences for customers. And then I always advise organizations to think about the long run, which is what's the new thing we can do? What's the new business model we can create or product that we can bring to the market? Uh That's something that should be on everybody's radar screen doesn't have to be today, but it's something you should be thinking about over the long run. Do you on these co pilots or are, are you uh thinking about developing industry specific co pilots, uh role specific co pilots? How, how far will Microsoft go with that? So through nuance, for example, which is a uh a company we bought last year, we're bringing some of the same co pilot capabilities into the health care space through assisting physicians. Uh We work, we'll be working with partners like PWC as an example to go into several industry verticals. Uh So yes, absolutely. We believe that co pilot model is going to be applicable to a variety of different industries. Much of that will work with partners to bring to bear because we're, you know, Microsoft obviously cuts across so many different spaces but and this is where expertise of different partners and different organizations we work with becomes I think really critical to developing the right tool set.

CEO Initiative 2023: Unlocking AI’s Potential…Securely

I would love to start with all of your perspectives on briefly sharing your view on the current state of generative A I and also the biggest risk that you think business leaders here today should know about. That's all. Well, I'll start, I think generally I, we heard this morning it's a general purpose technology that will impact every industry. I think the industries that will benefit and the companies that will benefit or be impacted. Uh Most in the short term are industries or companies that are digitally mature that have rich uh data sets and can move relatively quickly to take advantage of the new technologies in that spectrum. I think the industry that will accelerate probably the fastest unfortunately, is the cybercrime industry. Uh because cybercrime by definition is, is digital and it's largely actors who don't have to worry about, you know, ethical guidelines and safeguards and standards. And so they are able to move very quickly and we're seeing that already, we're seeing pretty much every company here is already being uh attacked using A I uh for fishing uh for, for malware, uh for social engineering with bots to try and find A way through, through your data and your application. So I think that will accelerate significantly with generating the A I with the ability to emulate humans and create much more intelligent social engineering and fishing attacks. So better than those emails I get about Alan needing gift cards, they're, they're going to get way better and way more intelligent. No, I think that what francois um outlined is, is, is probably the thing that in the short term is where the risk is. Um, but I'm actually pretty optimistic that we'll get past that. I think that the idea that we're verifying people by what we think are human traits. It's probably not a very good security implication in the first place like your banker shouldn't wire money just because you call up and you sound like you. And so I actually think that, yeah, that's gonna be a problem. We're gonna read some scary stories about, you know, parents that are tricked into wiring money to their, their kid who it sounds like is in, you know, some Mexican prison or something. But, you know, it turns out it was an attacker, you laugh but that that's gonna happen, happened recently and, and there's gonna be a bunch of those things and they're gonna make scary headlines. But I actually think it's going to accelerate us getting away from that idea that we can tell who to trust simply by how they sound or what they look like or those things. I think the longer term challenges for the CEO S that are in this room are gonna be much more around. How do they take advantages of what A I brings? But they do it in a way which is responsible over the long term. And so what we're starting to see is a lot of times employees at company are using A I to get their job done without understanding that the minute that they send that data off to an A I engine, it's gone for good, right? And it gets incorporated into it. A way of thinking it is imagine like a two year old and you accidentally swear in front of them. And then for the next period of time, all they're doing is just saying the bad word over and over and over again. A I is exactly like that you can't get the information back out of that. And that creates a real problem around trade secrets, that creates a real problem around data, locality and data compliance around the world right now, if you're using an A I system and you're a European based company, you are sending it all back to the United States and it's not even something where if you send a GDPR request and say, hey, forget that you can do it. It's embedded in a black box that is almost unforgettable. And so I think the thing that we're talking to our clients about is how do you make sure that as you're using A I, you're doing it anticipating that you wanna get those benefits, but at the same time, you wanna understand what the risks are and make sure that you're only letting the data flow out of your company. There's actually something you want to flow out of your company. Let's talk about the current state. Um I've been a leader of A I machine learning at some of the biggest companies in the world and A I, I've seen for the last decade drive truly digital transformation. Now, if it's the vehicle of trans of digital transformation, if anything generative A I is just speeding up that car on the highway and it's going super fast, what we are not doing are the basic maintenance, the basic safety and security checks on that vehicle. And so if I'm looking at where the risk is, the risk is really in the supply chain that we've seen, how are, how is A I being built? How are these models being built from the data, the foundational models, the open source libraries within the supply chain. So when we meet with customers, we ask two questions and I please take this back to your team. I ask how secure is your A I? And how do you know and how do you know is incredibly important because companies today don't have the capabilities or the ability to audit these systems and have visibility of their A I therefore, it's really hard to be able to secure it. Wow. I love that. Um Francois, you said that the number of cyberattacks on companies and applications have increased dramatically to the point where cyberattacks are costing companies $8 trillion a year globally. I found that number absurd, but in a crazy way, doesn't A I kind of arm the cyber criminals with tools that can now scale and what, what are do companies need to allocate more capital towards security? Well, yeah, the, the the cybercrime the cost, the cost around the world are around eight trillion, which is the third largest economy in the world after the US and China uh A I will accelerate that. Um And uh it, it does arm um cyber Attackers with more tools and therefore it will increase the speed and volume of attacks, not necessarily the sophistication that allows Attackers that would be not sophisticated at all to be able to launch an attack. One of the scary facts, for example is that today somebody that wants to attack, you would went somewhere doesn't actually know, doesn't actually need to know how to code that if you think about it. It is quite scary. Those folks can actually go online and you know, buy tools, there are tools like fraud GP T you can buy for $200 a month and gives you a starter kit for attacking companies with went somewhere and exploiting vulnerabilities immediately. Uh it doesn't mean that companies necessarily need to spend more money on, on cybersecurity. But it, it means that number one companies to be, need to be using the tools that exist today. I see a lot of companies that don't actually use the current generation of A I. There are a lot of tools today in security for security controls, for monitoring logs, for protecting against attacks that use the current generation of machine learning models to protect against these attacks. And too many times companies underestimate the sophisticated nature of their Attackers and therefore don't use those tools and get breached or under attack and don't even know it. And, and Ian uh what are some of the inherent risks uh That generative A I presents if those inputs in, in the safety belt isn't there uh when you're first starting. So a couple things as we think about generative A I and A I under the hood, it's, it's machine learning again. And when we talk about the differences and budgets that you brought up machine learning systems are different than all the other systems within your company. So as you're thinking about how do you protect your data, your infrastructure, your equipment, you need to think about how you're going to protect your A I. Now, what are some of those differences? The tools, the tools that developers and practitioners use to build models are different and today's security vendors are, don't scan those tools with a way to build its detect vulnerabilities. Also the models in the supply chain. Now, when you brought up generative A I, specifically, there's new types of attacks. One of them called prompt injection attacks. It's basically like a sequel injection attack where I can go in there and I can try to mislead the model to do something in a malicious way as a bad actor. Now, this becomes a lot more problematic when it's not just, you know, working with chat GP T to send me some information that you know, is crafting an email. It's when we start to connect these systems to agents, agents that carry out tasks. That's where the real risk lies. And therefore, that's why we need auditability, visibility and security of these systems. And Matthew, how are we seeing this practically uh at cloud flare? What measures are you taking to make sure that your company and the companies that you work with aren't victims of some sort of attack? Well, I think that the, the first thing and again to be the optimist on the panel is, you know, the whole thesis of cloud flare was that if we could get enough of the web behind us, if we could see enough information that we could feed that into what, what we frankly still are somewhat uncomfortable calling A I systems but are machine learning systems that could find new vulnerabilities and stop them before they were attacked so that you get the benefit of being part of the collective. That is, that is cloud flare. And what's been amazing in the same way that publicly we've all watched over the last 18 months as it seems like these A I systems went from jokes to, you know, writing sonnets, right? Which is pretty amazing. Um That same thing we've been watching internally where I remember about a year and a half ago, the first time our system found a vulnerability that no human had ever found before. And that went from something that was like, whoa, that's amazing to now, something that's happening on a very frequent and regular basis where we are oftentimes in front of these vulnerabilities and in fact, you know, have, have a number of those that were working with the industry in order to help clean up right now that we wouldn't have been able to find if it was simply humans that were out there. And so I think that the very value of this uh and the very value of using cloud services that you can uh have the benefits of all of the information, use that with the algorithms because the winners in the A I space aren't gonna, the algorithms are, are, are fairly, you know, trivial. They're, they're not, they're, they're the complexity, isn't there? The real thing that you need is to feed the algorithms with the right amount of data and in the security space every single day clouds or customers are safer because of the fact that being able to use these machine learning systems in order to protect everyone who uses our network. And, and again, that's where there are gonna be some scary stories that are out there. But I actually think that the good guys thankfully do have more resources than the bad guys. Um We, we have more access to data than, than they do. And there are ways that we're going to be able to not in the short term, short term is gonna be a little scary, but over the long term, I think the world becomes safer because of a I not, not scary. Have you ever gotten breached? Oh yeah, I mean, there's, we are constantly under attack. It's hard to imagine a threat actor that that isn't going after us. But what I really like is that the way that our systems are designed, we have an employee fall for a phishing scam, ok? And make sure that even when that happens, there's a belt and suspenders that keeps them from being protected and that is in part with machine learning system. So I'll give you a very specific example. I'm in Washington DC right now. Earlier this morning, I was in New York City. Um We have an A I system inside of cloud flow that is checking all of the our employees calendars to see where they are supposed to be and then automatically feeding rules back into the system where if I, um all of a sudden logged in right now to one of our systems, New York, it would simply, even if I had my user name and password and second factor authentication and used even my own machine because I've left it somewhere behind. That would still not let me through um our systems because we've been able to actually get smarter and more secure using the A I tools that we have internally and that we also then make available to all of our customers. Got it. Do we have any audience questions? No, everyone is sick of that. Um II I wanna ask Francois, how are you making sure that your workforce and the companies that you work with are up to speed and educated on protecting themselves from potential security risks? Well, so I think this is something every company out there needs to be doing, which is constantly training. Uh Our employees, Matthew mentioned earlier training, our employee against fishing and always tempting our, our employees with the fishing scam and see if they were doing that periodically or is it a one time training when they join the company? No, no, it's something that's on a constant basis. On a very regular basis. We train and retrain against these attacks. We analyzed employees who perhaps didn't get the training or fail for these attack and go back and, and train them on a very, very regular basis. That's something that we do today against regular attacks. But we're starting to train our employees on A I literacy because we feel that that's going to be something that every company needs to do to upscale their uh their employees and their knowledge to be able to both use the um uh A I in the most responsible way and understand best prac best practices for prompts. But also training uh engineers who are building these algorithms to know how to sanitize uh the data sets. Um And you know, be able to build these algorithms responsibly, you know, and I think that the thing though, that I would go even further on that is your employees are using A I, whether you want them to or not. And so the vast majority of employees are using A I and, and something like half are not telling their employers that they're actually doing it, they're using it for their, using it for their daily job to get to get work done. Because these are powerful algorithms. If you're having to write reports, you know, or you're having to generate code, it's a very safe place to or very productive place to be able to turn to what's again scary about it. It's a two year old. And so once you take your company secret and you send it to the two year old, it's gone forever, right? And that bad words gonna get said over and over and over again. And so I think that part of the tools here is not just in training, but actually putting in place a lot of the, you know, almost like D LP systems that say, you know, again, if you imagine there's a whole bunch of support documents, you're probably fine with those going to the A I systems and those learning. That's great because that actually is, is teaching the A is about how your product works. That's fine. You might have some other secrets that you say we can use on A I systems, but they have to be run locally inside of the company. And then there are other things, you know, that the, the sort of bad metaphor is the secret formula to Coca Cola. You may never want that to actually go into the, it's really tough because again, it's like the two year old, once these systems have learned something, it's not like a database row where you can go in and just delete it unlearning is actually this really fascinating area of A I research right now. And it's almost like a like AAA movie. It's like the, you know, turtle sunshine of the spotless mind. Like we don't have that system to have these, these A I systems unlearn things. And what that means is that it's actually a much bigger security threat for a company, even for insiders to make sure that as the data is leaving the systems, that it's the right data that's approved to be used. And that means training for sure, but it also means implementing the technical systems that restrict it and thinking about the supply chain, Ian, were you going to add something? Yeah, I think education is critical. I meet with 10 to 12 chief information security officers every week and I was meeting one with a big Pharma company this past week and he said, Ian, we're early in our A I journey. We're just now deploying A I. My response was no, you're not, you've been deploying A I for well over a decade as you look at drug discovery and every other facet of your business. But it's just now starting to get on to the table of the cyber security experts within the company. These systems again are different. We must educate how they're different and how we need to treat them differently in terms of looking for vulnerabilities, looking for exploits. And I would love to end on a very practical note. What are some tools that the business leaders here today can use to make sure that they're doing this responsibly and ethically other than promoting our own products? Um I think it's really just asking the simple questions and we heard it from Julie Sweet from accenture as well asking about who's in charge of responsible A I who's in charge of security of A I who's in charge of the visibility, the auditability of these systems just like your financial systems have a ledger. Your A I systems need to have a ledger. And oh, by the way, I only know five companies on the planet that I've worked with. I've worked with thousands that actually have that capability. That's where I would start. I, I'm not above promoting my own products, cloud players. A A I gateway gives you the tools to be able to control how data flows out of your systems. We're working with some of the largest companies in the world to make sure that they have control over that. We make it super easy for you to tag data and say here's where it goes and also control the budget that you have for your A I spend. Because the biggest thing that we hear is these budgets just blow up out of control in a way that even the cloud budget seem seem relatively limited. Awesome, great note to end on. Thank you to our panelists for joining us today.

Coursera

Coursera stock chart

Videos