Strengthening Cybersecurity Resilience in the Age of AI: Key Takeaways from the Hong Kong SFC Circular

The rapid proliferation of artificial intelligence (“AI”) across the financial services sector has introduced not only operational efficiencies but also a potent new generation of cybersecurity threats. Hong Kong Securities and Futures Commission (“SFC”) recently issued a circular calling upon all licensed firms to strengthen their cybersecurity measures against the emerging and escalating risks posed by AI-enabled cyberattacks (the “Circular”). This development arrives against a backdrop of sharply rising cybersecurity incidents in Hong Kong—a roughly 27% year-on-year increase to more than 15,000 recorded cybersecurity incidents in 2025, up from 12,536 in 2024—and a growing regulatory concern that frontier AI models are fundamentally reshaping the threat landscape.

Background and Scope

The Circular is addressed to licensed corporations, SFC-licensed virtual asset service providers, and their associated entities (collectively, “licensed firms”). The Circular draws upon observations regarding the preparedness of licensed firms for AI-assisted cyberattacks based on the SFC’s prior engagement with them and key internet trading platform providers, and distils these observations into practical guidance.

AI models can now plan and execute complex, multi-step actions autonomously, identify zero-day vulnerabilities, systematically chain multiple lower-risk vulnerabilities for high-impact disruption, and operate across interconnected systems to orchestrate large-scale attacks. AI tools also lower the technical barriers for phishing, social engineering, deepfake impersonation and reconnaissance, introducing new dimensions of speed and scale to the threat environment. As low-cost AI-enabled tools become more widely available, the volume and frequency of security patches and updates is likely to increase significantly, while the window between the identification of a vulnerability and its exploitation continues to narrow.

Senior management—in particular, the Manager-in-Charge of Information Technology (“MIC-IT”) at licensed firms—is expressly identified as bearing ultimate responsibility for managing the firm’s cybersecurity risk, and should ensure that changes to the firm’s cybersecurity framework and measures are properly and promptly reviewed, approved, and implemented. Where necessary, licensed firms should engage external IT security experts for advice and assistance.

The Circular

The Circular identifies five substantive areas for review and enhancement. An accompanying Appendix provides illustrative examples of controls for each area.

Patching and Vulnerability Management

Licensed firms should review and enhance patching processes, take prompt action on known vulnerabilities, and maintain policies for deploying urgent fixes outside routine cycles—particularly for business-critical components. The SFC also recommends decommissioning legacy systems, adopting risk-based prioritisation for patching work (considering severity, exploitability, external exposure, data sensitivity and criticality of affected system), automating patch deployment for lower-risk vulnerabilities where feasible, and implementing compensating controls (e.g., firewall or web application firewall) during the patch cycle to reduce exposure.

Access and Privilege Controls

Licensed firms should design controls assuming any user, device, privileged account or network component may be compromised, with access never implicitly trusted based solely on network location or user status. Recommended measures include enforcing least-privilege access to business-critical components, safeguarding privileged accounts, enhancing firewalls and implementing micro network segmentation to limit lateral movement, implementing maker-checker controls for high-impact actions, and preventing external inputs from altering systems instructions. The SFC also recommends phishing-resistant multi-factor authentication, least-privileged and just-in-time access, and strengthened perimeter controls to disrupt or delay attempted attacks.

Detection and Monitoring

Licensed firms should strengthen threat detection and monitoring of anomalies in client trading and system activities, and improve threat intelligence gathering. Recommendations in the Appendix suggest proactive scanning for systems, regular enhancement of anomaly-detection rules, periodic red-team exercises simulating AI-enabled and other cyber attacks, maintaining up-to-date threat intelligence through monitoring of news sources, and ongoing communication with security advisors and service providers.

Third-Party Supply Chain Risk Management

Licensed firms should implement procedures addressing AI-enabled threats targeting third-party service providers supporting critical operations, strengthen supply chain governance, and enhance service providers assessments factoring in the latest threat landscape. The SFC also recommends strong service providers due diligence referencing established information security standards and frameworks (such as ISO/IEC 27001), enhanced assessments of critical providers’ vulnerability management and patch deployment capabilities, assessment of concentration risk arising from reliance on a limited number of critical service providers, and contractual requirements for timely incident notification, vulnerability disclosure, and contingency arrangements.

Incident Response and Recovery

Licensed firms should establish and enhance their cybersecurity incident response plan and consider pre-planned containment strategies, such as blocking malicious activity, isolating impacted systems, and restricting access rapidly. They should test incident response procedures regularly through tabletop exercises or simulated attacks, and back up business records, databases and supporting documents regularly. Material cybersecurity incidents must be reported promptly to the SFC under the relevant code of conduct and guidelines. The SFC also recommends specifying roles and responsibilities of personnel, escalation protocols, pre-authorised containment actions, recovery actions, and communication strategies for affected stakeholders.

AI Language Model Risks: A Distinct Dimension

The Circular draws attention to the fact that a licensed firm’s own use of AI language models—whether internally developed, provided by the group company or a third-party vendor, or open source—may amplify existing cyber vulnerabilities or introduce new cyber risks, such as adversarial attacks against AI language models, data leakage, and system prompt override. Licensed firms should address these risks within their cybersecurity frameworks, consistent with the SFC’s November 2024 Circular on the Use of Generative AI Language Models, and comply with notification requirements for high-risk use cases under the Securities and Futures (Licensing and Registration) (Information) Rules. This signals that internal AI deployments form part of a firm’s attack surface and must be managed with the same rigour as any other firm systems.

Takeaways

The Circular reflects the SFC’s heightened supervisory focus on cybersecurity resilience amid an evolving AI-enabled threat landscape. While the guidance is directed at SFC licensed firms, the recommended measures may also serve as a useful reference for other organisations with operations in Hong Kong seeking to strengthen their cybersecurity frameworks against AI-enabled threats. We set out below some key takeaways:

1. Integrate AI governance with cybersecurity governance: The risks associated with the use of AI should be addressed as part of a broader cybersecurity management framework rather than treated as a separate compliance workstream. AI governance measures should be mapped into cybersecurity frameworks as part of a holistic approach to managing technology risk, with incident-handling procedures updated to address AI-specific vulnerabilities.
2. Technology Audit and Vulnerability Management: Organisations should consider conducting regular technology audits to identify vulnerabilities, assess the adequacy of existing controls, and ensure that their cybersecurity frameworks are aligned with regulatory expectations. In particular, firms with accumulated technical vulnerabilities or legacy systems should accelerate remediation timelines and ensure timely and effective patching.
3. Scenario-test for AI-enabled Attacks: Given the speed at which AI-enabled attacks can unfold, organisations should assess whether their pre-planned containment actions are sufficiently robust to address attacks that may compromise multiple systems within minutes. Tabletop exercises should incorporate AI-specific scenarios that account for the speed and scope of potential impact, and pre-authorised containment and remediation plans should be reviewed to avoid bottlenecks in a fast-moving incident.
4. Enhance Supply Chain Resilience: As AI-powered tools make it easier for threat actors to identify and exploit vulnerabilities at scale, organisations should assess their exposure to third-party cybersecurity risks and take steps to strengthen supply chain governance. This may include reviewing contracts with critical IT service providers to ensure they contain appropriate cybersecurity obligations, incident notification requirements, and audit rights to address the heightened risks. Organisations should also consider establishing clear protocols for information sharing when vulnerabilities are identified, and regularly test contingency plans to ensure operational continuity in the event of a vendor-related cyber incident.

The authors would like to thank Roslie Liu, Legal Practice Assistant at Mayer Brown Hong Kong LLP, for her assistance with this article.