Hong Kong’s SFC plans cybersecurity framework against AI-enabled threats

A common baseline and requirements proportionate to risk

Hong Kong’s Securities and Futures Commission plans to establish a cybersecurity framework for the firms it licenses. Chief Executive Officer Julia Leung announced the plans on November 2, 2026, at Hong Kong FinTech Week. The proposed approach would combine baseline requirements for all relevant firms with enhanced obligations for those with higher risk profiles. The official document sets out a future direction; it does not publish a new regulation already in force or a detailed compliance timetable.

The SFC links this preparation to intensifying AI-enabled threats. According to its published remarks, AI speeds up the transformation of hidden software weaknesses into attacks at scale. The regulator also says it is strengthening coordination with other Hong Kong authorities to share intelligence and improve financial resilience. It mentions joint efforts among Asia-Pacific securities regulators addressing operational incidents or scams that may cross borders.

Software alerts and a survey of twenty large groups

Leung cites recent global alerts involving PHP and other web development tools. She reports that some licensed corporations took websites offline as a precaution pending patches, while a few experienced disruption involving IT vendors. The document says the SFC’s own systems were not compromised because PHP patches were deployed early. These observations must therefore not be recast as an announcement of a regulator hack or linked to a particular new vulnerability without evidence.

She reiterates a June circular calling for critical fixes and sound cyber hygiene to be prioritized. She also presents findings from a survey of twenty large financial groups: 90% use or evaluate frontier AI for cyber defense, and nearly half use AI for adversarial simulation and threat hunting. The sample represents about 75% of securities turnover and 45% of clients reported by licensed corporations. These proportions do not describe every business in the market.

Agent autonomy must preserve human accountability

The same document says all surveyed groups already use AI in research, analytics, and back-office functions. More than half use or pilot agentic AI, and 80% anticipate further adoption over the next twelve months. A pilot and an intention must not be counted as widespread deployments. The SFC is developing practical, principles-based guidance for more autonomous applications, preserving human accountability and oversight throughout the workflow.

Leung distinguishes low-impact, reversible tasks from client advice, portfolio decisions, and market-facing execution, which need progressively stronger safeguards. Sandbox++ is intended to help examine how to reconstruct a decision, identify an action, and establish accountability. For financial intermediaries, the announced agenda thus combines cyber resilience with controls on autonomous uses. Final requirements and implementation arrangements remain the next points to verify in SFC publications.