Cisco (CSCO.US) has launched a security service designed to protect organizations from risks associated with artificial intelligence agents and other entities. By establishing trusted identities, enforcing strict zero-trust access controls, hardening agents before deployment, implementing runtime security protections, and providing Security Operations Center (SOC) teams with tools to block threats at machine speed, Cisco is integrating security into the foundation of the emerging AI economy.

For instance, OpenClaw, a recently popular AI agent platform highly regarded within tech circles, serves as an example. Developed by Peter Steinberger, who later joined OpenAI, OpenClaw enables users to configure their own AI agents. These agents can operate on a computer to perform various tasks, from checking emails and replying to messages to managing system files. While these capabilities are impressive, they also introduce potential security vulnerabilities, as users grant AI models permission to control their computers. While this enables super automation under normal circumstances, it could also lead to significant issues, such as permanently deleting important emails or entire programs from a system. This becomes an even greater concern for companies testing OpenClaw on corporate devices.

Consequently, many organizations are introducing new security products aimed at controlling platforms like OpenClaw to prevent their AI agents from causing more harm than good. Last week, NVIDIA (NVDA.US) unveiled its own such platform, NemoClaw, at the GTC conference in San Jose, California, while Cisco demonstrated its entry into this field on Monday.

"OpenClaw is to AI agents what ChatGPT was to chatbots; it essentially acts as a catalyst, making it easier for tech-savvy users to deploy agents in both workplace and home environments," explained Jeetu Patel, Cisco's President and Chief Product Officer. Patel added, "The downside, and the challenge... is the trust and safety risk associated with these agents. Because these agents... they are somewhat like teenagers. They are extremely confident, yet inexperienced, not fully aware of consequences, and unclear about the boundaries."

To address these concerns, Cisco stated it is adopting a multi-pronged approach. This strategy aims to protect the world from rogue AI agents, safeguard AI agents from external threats, and leverage AI itself to detect potential dangers from malicious actors and hackers. Describing this as "defending the world from AI threats" might sound exaggerated, but Patel believes it is warranted. AI, like any other software, can be exploited to attack infrastructure or steal sensitive data.

"The risks associated with these AI agents are far greater than those of chatbots," Patel noted. "The worst-case scenario for a chatbot is providing an incorrect answer. The worst-case scenario for an AI agent is the agent taking incorrect action, some of which could have catastrophic consequences, such as draining a bank account or deleting emails—actions that may be irreversible."

Cisco, along with other companies, believes the optimal approach is to treat AI agents similarly to human employees. This involves assigning them verifiable identities, then managing them and providing these digital assistants with appropriate access to the correct systems and services. Simultaneously, protecting AI agents from external interference is designed to prevent hackers from compromising an AI and forcing it to perform tasks it otherwise wouldn't.

Cisco announced that to achieve these goals, it is launching a service that will allow customers to test their AI agents' performance under simulated attacks. This will help identify their strengths and weaknesses. Finally, the company stated it is relying on AI agents to counter other AI threats, ensuring customers can respond to various scenarios at the same speed as potential attackers.

OpenClaw represents just the latest development in the rapid expansion of AI and is likely to drive further changes in the near future. After all, since its debut last November, it has already begun influencing business decisions at major companies. Naturally, this progression will necessitate more diverse types of cybersecurity technologies and methodologies to help defenders keep pace with attackers.