Protecting AI so AI Can Improve the World, Safely

The world is in the middle of an unprecedented era of artificial intelligence innovation. Looking ahead, there will be two types of companies: those who will lead on AI and those that risk irrelevance.

For the organizations who take AI seriously, the composition of their workforce is about to change dramatically.

Today, their workforce is entirely human. Tomorrow, it will expand exponentially to include a variety of AI workers—including apps, agents, robots, and even humanoids. We’ll be living in a world where connected AI agents and people work together to orchestrate all manner of complex workflows. And I believe it will translate into massive gains in productivity and capacity, with considerable shared benefits.

Imagine what a human population of 8 billion people can accomplish if we collectively have the capacity of 80 billion.

The question, though, is how can we make this transition safely and securely?

AI adoption introduces new risks

Keeping AI safe and secure in an enterprise is a difficult and relatively new problem. That’s because AI applications are built differently, adding a new layer to the stack: models. Unlike traditional applications, AI models can behave unpredictably, and the reality is that most organizations will be using multiple models across public and private clouds. This multi-model, multi-cloud and multi-agent landscape demands a new approach to safety and security.

Raising the stakes even more, when models fail, the consequences can be severe. Safety issues—like bias, toxicity, or inappropriate outputs—must be addressed, alongside threats from external actors exploiting vulnerabilities to steal confidential data or otherwise compromise your security. Model vendors and app builders will implement their own safeguards, but these measures while necessary will inevitably be fragmented and insufficient.

Ultimately, your security teams will need a common layer of visibility and control. They need to not only to see and understand everywhere AI is being used in your organization (by both users and app developers), they also must continuously validate and enforce your preferred guardrails over how AI models, applications and agents are behaving.

Introducing AI Defense: Reimagining safety and security for AI

You need to move fast with AI, but you absolutely cannot afford to sacrifice safety and security for speed. That’s why today, at our AI Summit, we announced Cisco AI Defense—a solution designed to eliminate this tradeoff and empower you to innovate fearlessly.

AI Defense provides robust protection in two critical areas:

1. Accessing AI Applications: Third-party AI apps can supercharge productivity but pose risks like data leakage or malicious downloads. With AI Defense, you gain full visibility into app usage and enforce policies that ensure safe, secure access—all powered by Cisco Secure Access and enhanced with AI-specific protections.
2. Building and Running AI Application: Developers need the freedom to innovate without worrying about vulnerabilities or safety issues in their AI models. AI Defense discovers your AI footprint, validates models to identify vulnerabilities, applies guardrails, and enforces them in real time across public and private clouds.

AI Defense is built on two game changing innovations we’re pioneering: continuous AI validation and protection at scale.

Validating at scale

You need to make sure your AI models are fit-for-purpose, and that they don’t have vulnerabilities, unexpected behaviors, data poisoning, or other issues.

For traditional applications, you’d use a “red team” of humans to try to break the application and find vulnerabilities. Unfortunately, this isn’t realistic for non-deterministic AI models.

This is where our AI Algorithmic Red Team capability comes in. It’s one of the big reasons why Cisco acquired Robust Intelligence last summer. They’re a team of AI security pioneers that have developed what we believe is the world’s first algorithmic red teaming solution.

The AI Algorithmic Red Team sends a successive series of prompt variants to a model to try to get it to provide responses it shouldn’t. Rather than having a red team of thousands of people try to jailbreak a model for weeks, we do it in just seconds.

It’s kind of like playing a game of 100 questions. But because it’s automated, it’s a game of 1 trillion questions. And AI makes 1 trillion look small.

Once AI Defense finds vulnerabilities, it recommends guardrails that you can apply. And it does this continuously. So, any time your model changes or any time there’s a new type of threat, your model is re-validated and updated guardrails are applied.

Protecting at scale 

Thanks to our platform approach, we can protect AI at scale in ways that only Cisco can deliver.

We already fuse traditional security directly into the network. You get thousands of distributed enforcement points, everywhere you need them, close to the users and close to the workloads. These control points can sit in an application in the public cloud, on the infrastructure in a private cloud, on a server, on a top-of-rack switch, or even out at the edge.

AI Defense takes full advantage of this platform approach so that your AI guardrails are likewise hyper-distributed and available wherever you need them. You get total visibility across your entire AI footprint, and the control to enforce everywhere.

Critically, AI Defense is also frictionless for developers. In fact, it’s invisible. There are no agents, it requires no libraries, nothing to slow down development. That means you can move fast to create new AI experiences and innovate for your customers.

Purpose-Built Technology Backed by Unmatched Intelligence

AI Defense is built on purpose-built technology and our own custom AI models powered by Scale AI. By working closely with leaders like Scale AI, and leveraging our own proprietary intelligence, AI Defense provides unparalleled insight, ensuring fast, efficient, and accurate protection.

Unlocking AI’s Full Potential

I’m incredibly proud of what our team has achieved with Cisco AI Defense. This solution empowers organizations to move fast, innovate boldly, and unlock AI’s full potential—securely and without tradeoffs.

Learn more about Cisco AI Defense and how it can protect your AI journey:

Watch the video

Read: Cisco AI Defense: Comprehensive Security for Enterprise AI Adoption

Share: