Flexxon wins third Singapore CyberCall grant to secure AI agents in hardware
What's the deal? Flexxon has been named an awardee of Singapore's Cybersecurity Industry Call for Innovation (CyberCall) 2025, its third award under the program. The grant, announced at CyberSG Innovation Day on November 10, 2026, will fund X-PHY AI ArmorDealroom has a profile for this one. Try Dealroom →, a hardware-rooted secure gateway for AI agents.
What's the endgame? X-PHY AI Armor will sit between an AI agent and the systems it reaches, inspecting every request for injected instructions and hijacked tools. High-risk operations will be handed to X-PHY hardware, which locks them down. Because authorisation is anchored in the device rather than software, a stolen credential alone will not get through.
Why now? The project targets CyberCall's "Cybersecurity for AI" challenge. CyberCall is run by the CyberSG Talent, Innovation and Growth Collaboration Centre (TIG Centre), a joint initiative of the National University of Singapore and the Cyber Security Agency of SingaporeDealroom has a profile for this one. Try Dealroom →.
How it works: Generic API gateways inspect traffic at the network layer and can be reconfigured by anyone who takes control of the host. X-PHY's enforcement sits below that: the device verifies authorisation and writes the audit record to hardware, so an attacker controlling the host cannot alter the decision or erase the trace.
"Agents are useful precisely because they act without waiting for a person, and that is exactly what makes them dangerous," said Camellia Chan, co-founder and chief executive officer of Flexxon. "You cannot solve that by asking software to watch itself more carefully — the judgement has to sit somewhere an attacker cannot reach."
What could go wrong? X-PHY AI Armor remains a development effort, not a deployed product with proven traction. Its value will depend on whether hardware enforcement protects more than the fast-growing field of software-based agent controls.
The signal: Security is shifting from protecting AI models and prompts to controlling what autonomous agents can actually do. Vendors including SentinelOne and CrowdStrike have added tools to monitor and restrict agent activity, and MCP gateways are emerging to govern how agents connect to external tools. Flexxon is betting that rooting those controls in hardware gives it an edge software alone cannot match.
Read more: pitchengine.com.au
Image credit: maushammer
