Enterprises are rushing to integrate AI agents and physical AI into their operations, but they are doing so on legacy network infrastructures that were never designed for machine-speed traffic, according to Cisco executives speaking at Cisco Connect 2026 Singapore.

In her opening address, Tay Bee Kheng, president of Cisco ASEAN, warned that the infrastructure most organizations have built for today is like a road network designed for bicycles. The comparison was meant to illustrate a fundamental mismatch: traditional enterprise networks were built for occasional human clicks, not for autonomous software agents that interact with applications, data, and other systems around the clock.

A network built for bicycles

Tay explained that a chatbot creates only intermittent pressure and demand on a network. A human user asks a question, waits for a response, and moves on. AI agents, by contrast, have sustained and persistent demand on enterprise infrastructure. They do not sleep, they do not pause for coffee breaks, and they often run multiple processes simultaneously. The network traffic patterns they generate are fundamentally different from those created by human employees.

“Infrastructure that is made for a human click on the network is not applicable for AI agents anymore,” she said.

She also offered a daunting perspective on the scale of the coming change: organizations must be ready to onboard as many as ten AI agents for every employee. That means an enterprise with 10,000 workers could need to manage 100,000 non-human identities, each with its own permissions, access rights, and behavioural patterns. “There’s no HR system for that at this moment,” she added.

The challenge extends beyond bandwidth. Legacy networks often lack the low-latency paths and dynamic routing capabilities that AI agents need to make real-time decisions. AI workloads increasingly depend on distributed computing, with models running across data centers, edge locations, and cloud environments. A network that was designed when traffic was predominantly north-south, from user devices to centralized servers, is poorly equipped to handle the east-west, machine-to-machine traffic that agentic AI creates.

Agentic AI creates security blind spots

With agentic AI capable of reasoning, planning, and acting across supply chains and operations, security teams face a new class of challenges. Unlike human employees, AI agents do not authenticate with a badge or a password in the traditional sense. They may use API keys, service accounts, or delegated credentials. They do not behave like humans either, making it difficult to detect anomalous activity based on user behaviour analytics.

During a media briefing, Robert Pizzari, group vice-president of Asia at Splunk, now a Cisco company, warned of the risks of shadow AI. Shadow AI occurs when employees deploy unsanctioned foundation models or give AI agents privileges they should not have. This can lead to data leakage, compliance violations, and unintended actions taken by autonomous systems.

Pizzari said shadow AI is inevitable. Rather than trying to ban it outright, Cisco is focusing on visibility and control. The company’s AI observability stack, bolstered by the recent acquisition of Galileo, gives enterprises the ability to monitor model drift and agent behaviour. Model drift refers to the gradual degradation of an AI model’s accuracy over time as the data it encounters changes. Agent behaviour monitoring tracks what an autonomous agent is doing, which tools it is using, and whether its actions align with the organization’s policies.

“Inevitably, shadow AI will be a feature and a function, so we also need to ensure that organisations have the ability to hit the handbrake,” Pizzari said.

Koo Juan Huat, Cisco ASEAN’s director of cyber security, stressed that treating AI agents with the same zero-trust architecture used for human employees is the only way forward. Zero trust is a security model that requires every user and device to be verified before being granted access, rather than assuming that anything inside the network is trustworthy.

Koo outlined three things enterprises will need to do to secure AI agents: know the agents on your network, know what they are authorised to do, and implement strict guardrails. He pointed to Singapore’s Government Technology Agency, which recently moved to build an AI agent registry for public officers, as an example of how a structured approach can help.

“You need to be able to look at what the agent is doing and give it permission just in time and just enough to do what it needs to do,” Koo explained. “A human needs to come into the loop and authenticate and authorise the action.”

Fighting frontier AI with frontier AI

As organisations deploy AI, so do threat actors. Rahayu Mahzam, Singapore’s minister of state for digital development and information, reminded attendees that AI-powered voice phishing attacks that cloned CEO voices in 2025 are no longer hypothetical risks. Those attacks demonstrated how easily generative AI can be abused to impersonate executives and trick employees into transferring money or disclosing sensitive information.

“Agentic AI is here – AI that doesn’t just respond, but reasons, plans and acts,” Rahayu said. “But with accelerating capabilities and automation come new digital and cyber risks. AI agents that act without sufficient oversight can cause real harm.”

To stay ahead of the threat landscape, Cisco is using frontier AI to defend the enterprise. Through Anthropic’s Project Glasswing, Cisco used frontier AI models, including Claude Mythos, to scan 1.8 billion lines of code across more than 25 programming languages in just eight weeks, achieving a false positive rate of under 3%. This kind of large-scale code analysis would be difficult, if not impossible, for human security teams to perform in such a short window.

But Cisco executives acknowledged that the industry’s security landscape is too fragmented. Different vendors use different formats for threat intelligence, and organisations often struggle to integrate security tools from multiple providers. To close the gap, Cisco is open-sourcing its AI security and safety frameworks to help the wider ecosystem build secure agentic AI systems.

These include DefenseClaw, an open-source framework that scans, sandboxes, and inventories AI agents, their skills, and their model context protocol (MCP) connections before they are allowed to run. Sandboxing is a security technique that isolates an application or agent in a controlled environment so that its behaviour can be observed without risking damage to the broader network. One of DefenseClaw’s components, CodeGuard, performs static analysis on agent-generated code to flag vulnerabilities before they can be exploited.

Upskilling people

Ultimately, the most advanced technology is only as trustworthy as the humans who govern it. At the event, Rahayu announced a three-year memorandum of understanding between Cisco and the Digital Defence Alliance Singapore (DDAS) to develop joint training programmes in AI and cybersecurity. The initiative aims to level the playing field in AI upskilling, creating practical learning opportunities for youths and working professionals.

Already, polytechnic students from the DDAS community are scheduled to visit Cisco’s Tokyo office in October to learn about network security in Japan’s commercial IT industry. The trip follows a similar visit to Seoul in April 2026. These exchanges are designed to give students hands-on exposure to real-world security operations and to build a talent pipeline for the region’s digital economy.

The broader technology industry is facing a significant skills gap. As AI reshapes the way software is built and operated, workers need new competencies in areas such as prompt engineering, model evaluation, AI governance, and secure software development. Training programmes like the one announced by Cisco and DDAS are intended to help bridge that gap.

Rahayu concluded with a challenge to the industry: “The question is no longer whether AI will transform the way we work. It already has. The question is whether we are ready to lead that transformation – with skill, with security and with trust at the centre.”

Source: ComputerWeekly.com News