Cisco Systems Inc is embracing a new approach to protecting artificial intelligence computing systems: using AI itself to keep them safe from attack.
The company, whose equipment serves as the backbone of computer networks, is introducing a service that relies on AI to automate security tasks, such as patching vulnerabilities, handling testing and deploying upgrades. It also can isolate an intrusion when it happens.
Beefing up cybersecurity is more important than ever as data centers refocus their computing power on developing and running AI models, according to Cisco Chief Executive Officer Chuck Robbins.
“The data centres that power the AI revolution are going to be completely different in the future,” he said in an interview. “They’re going to require new infrastructure, new ways of processing data and, obviously, deep levels of security.”
The new Cisco service, called Hypershield, represents the biggest overhaul of its security offerings to date. It’s also an attempt to play a bigger role in the AI boom that’s helped turn companies like Nvidia Corp. into Wall Street darlings.
The push fits with Robbins’ longer-term effort to transform Cisco into more of a services company that makes money from subscriptions. Hypershield allows a broader range of customers to have the level of sophistication normally seen only at the largest cloud computing providers, Cisco said.
The new service doesn’t completely hand over control of security to machines, at least not initially. The technology starts off using automated recommendations that humans can oversee – until there’s enough confidence to let the software take more complete control.
There’s a reason AI systems require a different approach to security, said Cisco Executive Vice President Jeetu Patel. Instead of humans simply accessing information from remote computers, there is a much more complicated network of devices that communicate with one another. There also may be numerous remote AI assistants for each user, all accessing information and interacting in different ways.
Hypershield aims to handle mundane tasks that take too long and leave networks vulnerable. Currently, when a weakness is found, more than 45 days can pass before a software patch is created, tested and installed. The breach, meanwhile, can allow entry into exposed systems within three or four days, Patel said.
Once hackers make their way into a network, they typically move around inside – jumping between elements. Hypershield is designed to monitor and segment the network, limiting the access a bad actor might be able to get.
The new product will automate the time-consuming process of upgrades as well. It will handle that by testing updates on a digital twin of the user’s network, making sure it works as intended before installation. Hypershield also will work with Nvidia’s AI Enterprise service, which includes AI-driven anomaly detection. – Bloomberg