Skip to content

Invest1 publisher3 min readPublished

Hong Kong's SFC chairman puts AI-assisted decisions under Section 465 director duties

Kelvin Wong told a Hong Kong governance conference that boards own AI-assisted decisions. Almost nine in ten listed issuers mention AI in their reports and fewer than a fifth describe a framework for governing it.

The Investor · Invest desk

Illustration accompanying Hong Kong's SFC chairman puts AI-assisted decisions under Section 465 director duties

What happened

  • SFC Chairman Kelvin Wong told the Corporate Governance Conference 2026 that AI governance is a board-level responsibility and that directors keep human accountability for AI-assisted decisions.
  • Wong cited a market of roughly 2,700 listed companies in which about 9% account for more than 90% of average daily turnover.
  • In June the SFC told licensed corporations, licensed virtual asset service providers and associated entities to strengthen their defenses against AI-enabled cyberattacks.

Compiled by The InvestorSomething wrong?How this is made

Why it matters

  • exposure Wong wants controls to match the statements issuers made to investors, so an AI paragraph already sitting in an annual report becomes the standard the company gets measured against.
  • decision Approving further AI spending now means naming expected returns and the conditions that would end the investment, which puts a revisit date on budgets that were signed without one.
  • constraint Basic AI literacy for every director, with a higher bar for whoever holds the technology or cybersecurity brief, narrows the pool of people who can credibly sit on a board running AI systems.
  • precedent The SFC's written AI expectations so far run to licensees on cyber defence, leaving issuer-level AI governance in speech form until a rule or a case follows.

Almost nine in ten of the more than 2,500 Hong Kong-listed issuers in the Hong Kong Chartered Governance Institute and Wizpresso sample referenced AI in their annual or ESG reports, and fewer than 20% disclosed a structured governance framework covering oversight roles, policies and lifecycle controls [5][6]. On a base of 2,500, that is about 2,250 issuers telling investors AI is in the business and fewer than 500 telling them who inside the company owns it, a gap of roughly 1,750 companies [1][2][3]. The study named its weakest disclosure areas as risk, compliance and incident management, AI governance and responsible use, executive ownership, and AI-related ESG issues [7].

Wong asked boards to understand the AI use cases inside their companies, assign named owners, and make controls match the statements made to investors [9]. The third item is the one an outsider can check. "Functions may be delegated; responsibility cannot," Wong said, referring to directors' duties under Section 465 of Hong Kong's Companies Ordinance [10]. He said all directors should have basic AI literacy, and that directors with technology, data, cybersecurity or AI responsibilities would face higher expectations based on their relevant expertise [11].

What the SFC has put in writing to firms this year is narrower in scope. In June it told licensed corporations, licensed virtual asset service providers and associated entities to strengthen defenses against AI-enabled cyberattacks, through updated technology inventories, vulnerability management, access controls, third-party risk management and incident response [12]. The regulator said frontier AI models could reduce the expertise, cost and time required to identify and exploit software vulnerabilities [13]. Those instructions bind licensees, while the speech addresses the boards of roughly 2,700 issuers [6].

Hong Kong has about 2,700 listed companies, and roughly 9% of them account for more than 90% of average daily turnover, according to figures cited in the speech [14]. That works out at about 243 companies carrying the turnover, with about 2,457 sharing what is left [4][5]. Among issuers with market capitalizations below HK$4 billion, about 40% have a shareholder controlling more than half the shares [15]. Wong said such conditions can increase the importance of independent information and oversight, and that AI could help investors identify anomalies in corporate disclosures while also amplifying selective narratives, misinformation and herd behavior in less liquid parts of the market [16].

On spending, Wong asked boards to assess what business problem an investment addresses, the expected returns and risks, and the conditions that would prompt the company to expand, change or terminate it [17]. "Activity is not achievement. Expenditure is not value creation. And adoption is not transformation," Wong said [18]. The termination condition is the demanding item, because a board that writes one down has to come back to the spend on a date.

In my view the first response will be disclosure language, because another paragraph in an annual report costs less than a named owner with a budget, and the study shows which option most issuers have taken so far [6]. The counter-case is that Section 465 attaches to individuals, and Wong said directors holding the technology, data or cybersecurity brief will be judged against that expertise rather than the board average [11]. That puts a person in front of a bad AI-assisted disclosure, not a committee. If the share of issuers disclosing a structured framework moves well clear of 20% in the next reporting cycle, the speech was enough on its own; if it sits at the same level, the next step falls to a rule [6].

What to watch

  • Any SFC enforcement action or listing rule consultation that cites director duties in an AI oversight context.
  • Whether the SFC extends its June AI-enabled cyberattack expectations from licensees to listed issuers.
  • Whether issuers begin publishing the expand, change or terminate conditions Wong asked boards to set for AI investments.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories