00:00:02 · G2 Patel, Tom Keene, Paul Sweeney, S3
[00:00:02] S0: Bloomberg Audio Studios. [00:00:04] S1: Podcasts, radio, news. So this is the basic idea. Switches, routers, wireless networks, data center, AI fabric. [00:00:14] S2: How about that? It must be Cisco Systems. Good morning, Mr. Chambers, if you're listening. [00:00:18] S1: Jeetu Patel is Chief Product Officer at Cisco. [00:00:22] S2: The Royal Bank of Canada says we don't trust our walls. [00:00:27] S1: Is the next step in AI this massive security risk that Paul Sweeney's checking account is unclogged. [00:00:35] S0: It is, Tom, the central issue of our time at this point because these agents that we are moving into, so we're in the second phase of AI where you've got this moment just like chat GPT that's happening with personal agents. The difference between a chatbot and an agent is a chatbot, you just ask it a question, get an answer. With an agent, you actually give it tasks and jobs it can conduct on your behalf fully autonomously. And one of the biggest challenges you have with agents right now are people are concerned about how much you can trust these agents because there's a paradox of a sort because with agents, they're not that useful if you don't give it access to your tools, your email, your calendar, your credit card. But the moment you give that access, your risk goes up quite dramatically. And so we have to make sure that we address the trust, safety, and security issue. [00:01:24] S2: Okay. [00:01:25] S3: How do we do that? [00:01:26] S2: And who does that?
00:01:26 · Paul Sweeney, G2 Patel, S3
[00:01:26] S2: And who does that? [00:01:28] S0: Well, we are spending billions at Cisco. Cisco is one of the largest security and safety companies in the world, and we're doing that. But the way you should think about how we do it is pretty important. The difference is that these agents are like teenagers. They're supremely intelligent. They have no fear of consequence, and they exercise poor judgment far more frequently, Paul, than you and I would like. Now, the way that you have to think about this is, It's going to get harder and harder to distinguish when there's a malicious attack that's altering the behavior of an agent versus an agent just exercising poor judgment because it's very fastidiously just following a goal that you gave it to do. And so the way that you have to do it is you have to monitor an agent's behavior continuously. And when you start seeing drift and misalignment between human expectations and the agent's behavior, you have to be able to intercept right away. And so there's a whole set of technologies around that. And then the other area that people don't talk about as much is this notion of your ability as an organization to efficiently manage and use tokens is directly proportionate to your security posture, which means the lower and the more efficient you are with using tokens, the better your security posture. [00:02:51] S3: Where is the average corporate American enterprise with AI just in general, including using it wisely, protecting its safeguards, all that. [00:03:02] S2: Where are we today?
00:03:02 · Paul Sweeney, S3, G2 Patel, Tom Keene
[00:03:02] S2: Where are we today? [00:03:03] S3: It feels like we're super, super early. [00:03:05] S0: We're very early. I think what has started getting diffused is a lot of people within the enterprise now use chatbots just like they use. [00:03:15] S1: The search engine. [00:03:16] S0: So that's happening at scale. But when you start thinking about agents, we're still in the very, very early days. And I think part of the reason for that is just people being fearful of trust. [00:03:26] S2: Jeetu [00:03:26] S1: Patel, have you put your checking account on Muse? [00:03:30] S0: I have not put my checking account on Muse, but I use Muse regularly. The reason I haven't put my checking account on Muse is that's the area that I feel like we need to make sure that we're actually being careful before we go out. [00:03:43] S2: And give too much access. Fancy Cisco phrases. John Chambers would have never used these languages. Ancient hallucinations. Alert fatigue. Translated, we don't trust the computer. That's all there is to it. How does your industry, with great respect to Mr. Chambers and all that he invented and sustained, I should say, across some tough years, how do we get the trust so we do transact given hallucinations and fatigue? [00:04:11] S3: I think.
00:04:11 · S3, G2 Patel
[00:04:11] S3: I think. [00:04:12] S0: What ends up happening in this market is you have to make sure that the safety and security technology is moving and innovating as fast as the innovation in AI. And once that starts to happen, what you start to see is the comfort level goes up with people. And that's just, it's a matter of time. And, you know, I was just with some of the other folks on a TV show over here with Bloomberg, and they were asking me, like, who's going to pay for this? Well, I think The revenue is there. If you look at just what Anthropic revenues are, that's the end user revenue that's being paid for. And then if you look at what, you know, both Azure and Google reporter has revenue on the cloud, that shows you that the infrastructure is also getting paid for. What you need to have is the trust and safety and security being, once that goes up a notch or two, you'll start to see a massive kind of upsurge that happens within usage. [00:05:07] S3: I'm going to say maybe two years ago, I was speaking with a management consultant and he was saying, His clients every day are calling in saying, this is two years ago, how do I implement AI into my business? Is it a friend? Is it an enemy? And he says, before we engage anybody, we have them sign a document that says their C-suite, their board, commits to implementing AI on a morally just basis and with all the guardrails and things like that. That kind of blew me away. But now it seems like that's kind of what you have to do. Are you seeing corporate America think about it that way?
00:05:07 · S3, G2 Patel
[00:05:07] S3: I'm going to say maybe two years ago, I was speaking with a management consultant and he was saying, His clients every day are calling in saying, this is two years ago, how do I implement AI into my business? Is it a friend? Is it an enemy? And he says, before we engage anybody, we have them sign a document that says their C-suite, their board, commits to implementing AI on a morally just basis and with all the guardrails and things like that. That kind of blew me away. But now it seems like that's kind of what you have to do. Are you seeing corporate America think about it that way? [00:05:40] S0: You have to make sure that the way that the guardrails and the value system that your organization has gets implemented. And that is a prerequisite for success for enterprises, especially the regulated ones. And so that's not really that controversial or debatable anymore. That's what everyone does. The question is how. And how do you implement them? Because these agents and these models tend to be unpredictable, where you don't really know how they're going to behave. And so you have to make sure that there's a constant monitoring. And the moment you start seeing drift and misbehavior in some ways, you have to make sure that there's implemented guardrails that actually stop it from doing something you don't want it to do.
00:05:40 · G2 Patel, S3, Tom Keene
[00:05:40] S0: You have to make sure that the way that the guardrails and the value system that your organization has gets implemented. And that is a prerequisite for success for enterprises, especially the regulated ones. And so that's not really that controversial or debatable anymore. That's what everyone does. The question is how. And how do you implement them? Because these agents and these models tend to be unpredictable, where you don't really know how they're going to behave. And so you have to make sure that there's a constant monitoring. And the moment you start seeing drift and misbehavior in some ways, you have to make sure that there's implemented guardrails that actually stop it from doing something you don't want it to do. [00:06:21] S3: So when you see a company like Anthropic, file an S-1, and then the risk section is the world's going to come to an end, AI is going to destroy everything. I can't even imagine being the lawyer of the banker who was tasked with writing up that silly risk factor, but it's not silly, is it? [00:06:38] S0: Well, I think there is a considerable risk of things going wrong. I think we have to make sure that we're careful on not being hyperbolic about it either. So while there might, two conditions can hold true. You can have risk and humanity does not need to come to an end. [00:06:58] S1: Do you see a roll up here? You mean if there's going to be transactions and combinations down the line, success, failure, how do you see that unfolding into anything but monopoly or even monopsony characteristics?