跳转至

Cisco Preps For A World Of AI Agent Coworkers, Frontier Model Threats

Ch01.035 Cisco Preps For A World Of AI Agent Coworkers, Frontier Model Threats

📊 Level ⭐ | 8.8KB | entities/cisco-preps-for-a-world-of-ai-agent-coworkers-frontier-model-threats.md

Cisco Preps For A World Of AI Agent Coworkers, Frontier Model Threats

URL: https://www.nextplatform.com/ai/2026/06/03/cisco-preps-for-a-world-of-ai-agent-coworkers-frontier-model-threats/5250406

深度分析

1. AI Agent 作为同事：企业基础设施的准备

Cisco 预见 AI agent 成为"数字同事"——企业需要为 agent 准备身份管理、权限控制、审计日志等基础设施，就像为新员工准备工位一样。

2. Frontier model 的安全威胁

前沿模型的能力提升带来新的安全威胁——更强的模型可能被滥用于社会工程、代码漏洞利用和信息操纵。

3. 企业网络的 agent 准备度

企业网络当前设计假设所有流量来自人类用户——agent 流量有不同的模式（高频、自动化、批量），需要不同的监控和安全策略。

实践启示

1. 为 agent 准备"数字工位"

身份、权限、审计、网络——为 agent 准备与人类员工同等的基础设施。

2. Agent 流量需要独立的监控策略

Agent 流量的模式不同于人类——需要独立的异常检测和安全策略。

3. Frontier model 的安全评估

追踪前沿模型的能力评估和安全测试——了解最新威胁模型以更新防护策略。

相关实体

→ 原文存档 - olmo-eval: an evaluation workbench for the model development

摘要

Published Time: 2026-06-03T00:10:37.000Z

Markdown Content: Last month, we did a deep dive into Cisco Systems’ third quarter fiscal 2026 numbers to get a gauge on what the AI market looks like when you strip away what the hyperscalers, cloud builders, and AI model makers are spending for their AI datacenter buildouts and instead focus on the up-and-coming AI users – think enterprises, neoclouds, and sovereigns – and the traditional hardware vendors that will supply them.

Those include system builders like Dell – and we heard a lot about what the company is doing with AI last month during the Dell Technologies World 2026 show – and later this month, HPE will weigh in at HPE Discover 2026 in Las Vegas. This week it’s Cisco at Cisco Live 2026, also in Vegas. Cisco made a strong push into the AI field at last year’s event, and this year they further outlined the latest efforts to build out its AI infrastructure stack for enterprises, particularly in this still emerging world of agentic AI – and now frontier AI models – and the challenges they present.

NEXTPLATFORM AD

Those challenges touch on all aspects of IT, from computing and networking to security and data. For example, AI agents create a tremendous amount of traffic. In its recent report, AI Impact on Wide Area Networks, Cisco found that agents generated 450 percent more traffic per task than with traditional tools. In the realm of data security, AI and automation drastically reduced the average breakout time – the amount of time between an initial network compromise and the moment attackers begin moving laterally through the network – to 29 minutes. And the introduction of frontier AI models like Anthropic’s much-talked-about Claude Mythos Preview could drive that down even more.

The broad reach AI has throughout an enterprise has Cisco, like its competitors, putting pieces in place to ensure its technologies reach every area, and that was on display during the first day of Cisco Live. With each announcement, executives wanted to show a fully integrated stack of Cisco networking, networking, and security for datacenters and cloud environments that can draw in tools from tech partners, drive faster deployment and implementation through validated designs and pre-configured designs, and include security capabilities that are increasingly important as AI becomes embedded in infrastructure and agents take on business operations.

A ‘Very Unique Advantage’

The vendor has been building out its infrastructure, software, services, and AI capabilities for several years, which DJ Sampath, senior vice president and general manager of Cisco’s AI Software and Platform Group, said gives the company a “very unique advantage” over competitors.

NEXTPLATFORM AD

“We manufacture our own silicon and that gives us the ability to start to reimagine a co-designed full stack that goes all the way from silicon to the network to the security being fused into that network in a way that nobody else can then combining that with the ability to leverage our data platform that brings all of these telemetry together and finally making it incredibly simple to operate and effective across these different domains using what we call AgenticOps,” Sampath told journalists and analysts.

A key to what Cisco wants to do lies in the understanding that as AI agents become more integrated into business operations, the operating model has to change and be simplified to create the conditions in which humans and agents can work together.

“One of the big transformations that's happening is we're starting to see this evolution happen from chatbots to agents, and when we talk about agents, what we're starting to see is agents are sort of manifesting themselves as additional coworkers,” he said. “You're onboarding them, you're working with these agents and these agents are starting to become very intrinsically woven into how you're starting to think about your day-to-day work. It's no longer about humans clicking through dashboards, a multitude of dashboards, and trying to keep up with what the agents are doing. A true collaborative operating model starts