On August 19, 2026, the Hong Kong China Network Security Association hosted “The Cyber Executive’s Guide to AI Governance and Risk” luncheon in Hong Kong. Industry representatives from the financial services, technology, legal and risk management sectors gathered to exchange insights on AI governance, data risk management and AI security practices.

The event was jointly organised by the Hong Kong China Network Security Association, AIStorm, together with leading multinational consulting organisations and cybersecurity vendors. It aimed to deepen industry understanding of the governance challenges and risk-management requirements arising from AI adoption, while exploring how organisations can build secure, compliant and trustworthy AI environments.

AIStorm played an active role in the event. Wu Xiangning, Senior Vice President and Chief Customer Success Officer of AsiaInfo Security, delivered a keynote titled “From Boxes to Systems: The Age of AI Agents — Securing the Next Generation of Digital Workers.”

His presentation examined the paradigm shift in enterprise cybersecurity and cost management as AI agents become increasingly embedded in business operations, and introduced innovative solutions designed to address both security and cost management.

From Fragmented Tools to System-Level Enablement: Rethinking Security Governance for Digital Workers

During the event, Wu said:

“Enterprise cybersecurity has long been trapped in a cycle of buying new boxes every year and stacking tools together. Traditional network, endpoint and cloud security tools operate in isolation—like a football team whose players never pass the ball—making it difficult to respond effectively to increasingly complex threats.”

Drawing on its cybersecurity capabilities, AI expertise and experience serving key industries, AIStorm highlighted the need for enterprises to move beyond fragmented security tools and build an integrated system encompassing Find, Block and Monitor, enabling coordinated governance across different security scenarios.

At the same time, as large language models and AI technologies penetrate core business operations, enterprises face an entirely new set of challenges. The ratio of AI agents to human employees within enterprises has reportedly reached 45:1 and continues to grow rapidly.

AI agents are no longer simply automation scripts. They are becoming “digital employees” capable of independently accessing systems and executing decisions.

Confronting Two Emerging Risks: The Invisible “Security Storm” and the Compute “Black Hole”

While enterprises are benefiting from the productivity gains created by AI, they are simultaneously facing two unprecedented challenges:

Invisible security blind spots: Research cited in the presentation indicates that 68% of enterprises cannot distinguish whether actions within their systems are performed by humans or AI. While 71% already allow AI to interact with core business systems, only 16% have precise controls over AI permissions. As a result, 88% have reportedly experienced agent-related security incidents. Uncontrolled API keys and excessive execution privileges are turning AI into a potentially vulnerable entry point within enterprise networks.

Uncontrolled AI compute costs: As AI agents make intensive use of large language models, token consumption can become an invisible cost centre. Most enterprises lack the ability to attribute AI compute expenditure accurately or route tasks intelligently between different model tiers. This can result in expensive models being used for relatively simple everyday tasks. Unnecessary compute consumption directly undermines the business return on AI investments.

AIStorm: A Dual Approach to Security + Compute Governance

To address these challenges, AIStorm presented a dual-solution approach designed to provide enterprises with a secure and economically sustainable foundation for digital productivity in the agentic era.

Agent Security — A Dedicated Security Framework for AI Agents

AIStorm's Agent Security approach provides lifecycle protection specifically for AI agents. It includes:

Discover — identify potentially hidden agents and establish visibility over AI assets.
Passport — provide each AI agent with a digital identity credential.
Manage — enforce strict least-privilege access.
Protect — intercept malicious or abnormal behaviour at machine speed.
Audit — maintain comprehensive, traceable records of agent behaviour.

Token ERP — Token Resource Management and Optimisation

Token ERP applies traditional ERP principles to the management of large-model compute resources. It enables enterprises to:

Manage — gain detailed visibility through management dashboards.
Control — establish budget quotas by team and AI agent.
Optimize — intelligently route workloads to the most cost-effective model for each task.
Save — reduce AI compute costs while maintaining business quality, maximising the value of limited AI budgets.

Strengthening Enterprise Security in the Age of Intelligent Connectivity

Concluding his presentation, Wu said:

“In the past, cybersecurity was fundamentally about protecting humans from malicious software. In the age of interconnected AI agents, our mission is to help enterprises manage and protect agents that think and act like humans.”

As enterprises worldwide accelerate deeper into AI adoption, AI governance is no longer simply an IT issue. It is becoming a core capability that can determine the success or failure of an organisation’s digital strategy.

AIStorm will continue to build on its expertise in cybersecurity and AI, deepen collaboration with partners in Hong Kong and globally, and develop more systematic and granular solutions to help enterprises securely navigate their intelligent transformation.