Flexxon wins funding to build hardware-based security for AI agents

The hardware-rooted solution

Flexxon has been awarded funding under the Cybersecurity Industry Call for Innovation (CyberCall) 2025 to develop a new security tool called X-PHY AI Armor. This project was announced at the CyberSG Innovation Day on November 10, 2026, as part of the "Cybersecurity for AI" challenge. The initiative is organized by the CyberSG Talent, Innovation and Growth Collaboration Centre (TIG Centre), which is a joint effort by the National University of Singapore and the Cyber Security Agency of Singapore.

The goal of X-PHY AI Armor is to create a hardware-rooted secure Model Context Protocol (MCP) gateway. This technology will sit between an AI agent and the various systems or databases it needs to reach. By operating at the hardware level, the tool intends to close the security gap that exists when companies rely solely on software-based controls to manage AI behavior.

This marks the third time Flexxon has received funding under the CyberCall initiative since 2019. While their previous projects focused on securing data at rest and server security modules, this new project addresses a modern problem. The company is leveraging its existing expertise in hardware cybersecurity to build a solution for the rapidly growing field of agentic AI.

Closing the software gap

AI agents are designed to perform tasks autonomously by holding credentials to read files, query databases, and call various tools. Because they act without waiting for a human to check each step, they can be easily manipulated if they receive poisoned instructions. A single malicious line of text in a document or a web page can redirect an agent to perform actions that remain technically authorized but are actually harmful.

Current security models often struggle because the controls meant to supervise these agents run on the same software layers that an attacker might try to take over. If an attacker gains control of a host, they can often reconfigure generic API gateways or delete audit records. X-PHY AI Armor aims to solve this by moving the enforcement layer below the software level where a host-level attacker cannot reach it.

The system will inspect every request an AI agent makes and look for hijacked tools or injected instructions. If the system detects a high-risk operation, it will hand the task to the hardware to lock it down. Because authorization is anchored in X-PHY hardware rather than software, a stolen credential alone will not be enough for an attacker to bypass the security measures.

The scale of the AI threat

The rise of agentic AI has moved from simple demonstrations to active deployment in just the last two years. However, the security models governing these systems have not kept pace with their adoption. The Cyber Security Agency of Singapore warned in a recent report that these systems could automate significant parts of cyber kill chains by chaining exploit steps together autonomously at high speeds.

This automation allows intrusions that used to take days to unfold in just a few hours. The report specifically cited OpenClaw, an open-source agentic AI technology, as an example of how these tools were abused to breach development pipelines at scale. As these agents begin to hold more sensitive credentials, the risk of automated, high-speed attacks on infrastructure becomes a primary concern for security professionals.

Market trends suggest that the integration of these agents will accelerate quickly across the corporate world. Gartner forecasts that up to 40 per cent of enterprise applications will ship with task-specific AI agents integrated by the end of 2026. This is a significant jump from fewer than 5 per cent just one year earlier, highlighting the urgency of developing robust security frameworks for these autonomous tools.

Impact on high-risk sectors

The organizations most likely to feel the impact of these security challenges are those that give AI agents access to sensitive data. This includes the financial services, healthcare, and government sectors, as well as the data center operators who run the underlying infrastructure. These industries require high levels of integrity and are prime targets for the types of automated exploits that agentic AI can facilitate.

Flexxon's solution is intended to help these organizations harness the benefits of AI agents while safeguarding their data. By providing a way to verify authorization for high-risk operations through hardware, the tool ensures that even if an agent is tricked, the damage is contained. The audit records for these actions will also be written directly to the hardware, making it impossible for an attacker to erase their tracks.

Ms. Camellia Chan, the Co-founder and CEO of Flexxon, noted that agents are dangerous because they act without waiting for a person. She explained that you cannot solve this by asking software to watch itself more carefully. Instead, the judgment must sit in a place where an attacker cannot reach, which is the core principle behind the X-PHY AI Armor project.

Future of AI security

The development of X-PHY AI Armor represents a shift toward hardware-anchored ecosystems for AI. As organizations move toward more complex agentic workflows, the need for