Skip to main content
Welcome to Deloitte

If we have selected the wrong experience for you, please change it above.

Frontier technology regulatory updates in financial services, quarter three 2026

 

The Deloitte Asia Pacific Centre for Regulatory Strategy (ACRS) is pleased to share the first edition of frontier technology regulatory updates in financial services. This publication brings together selected developments in artificial intelligence (AI), cyber resilience, digital identity, digital assets and quantum computing affecting financial institutions across Asia Pacific (AP) during the third quarter (Q3) of 2026. It covers both financial-sector measures and wider policy, legal and regulatory developments relevant to the industry.

Key developments this quarter included:

1. AI governance and controlled experimentation

Legislative proposals in both Malaysia and the Philippines share an emphasis on establishing mandatory, risk-based AI governance frameworks under central regulatory oversight. Malaysia’s National AI Office consulted on a proposed AI Governance Bill that would introduce a central authority and a risk-based framework including an ‘unacceptable’ risk category. The Malaysia Securities Commission also consulted on board oversight of technology and disclosures on the use and governance of technology and AI by publicly listed companies. In the Philippines, the proposed Artificial Intelligence Regulation Act would establish a National AI Commission and mandate system registration while also requiring developers and deployers to implement mechanisms for immediately halting any AI system that autonomously exceeds its intended function. Alongside this legislative track, the Bangko Sentral ng Pilipinas (BSP) issued its voluntary Governance Principles for Artificial Intelligence in Financial Services. These principles extend to third-party vendors and outsourced partners under a shared-responsibility model.

Australia announced plans to introduce national AI standards to regulate large data centres and frontier AI training, alongside establishing an office of AI. Under the proposed tiered framework, large data centres would face mandatory requirements governing energy efficiency, renewable sourcing, sustainable water consumption and workforce skills and training. Meanwhile, frontier AI labs authorised to conduct large-scale training in Australia would be subject to proposed minimum security and safety expectations, including the mandatory reporting of defined AI incidents. Alongside this infrastructure focus, the government outlined broader consumer safety priorities, including a proposed digital duty of care, privacy reforms, workplace safety measures and the examination of consumer-law options to address risks arising from automated, agentic transactions. On the market side, the Australian Securities and Investments Commission (ASIC) amended its Market Integrity Rules to strengthen safeguards for algorithmic and AI-enabled trading, including controls enabling the immediate suspension, limitation or prohibition of trading involving particular algorithms. The amended requirements will commence on 18 March 2028. Japan’s second AI Basic Plan addresses the expansion of agentic AI through design safeguards, improved AI literacy and a strategic focus on ‘AI sovereignty’ to avoid excessive dependence on any particular country or company.

This focus on the systemic risks of autonomous systems is mirrored in Taiwan, where the Ministry of Digital Affairs (MODA) launched its AI Risk Taxonomy and Assessment Framework (v1.0) to serve as a common standard for sectoral regulators, including the Financial Supervisory Commission (FSC), to draft industry-specific rules. The advisory framework groups 20 risk sub-types into technical, operational and societal categories. Under the operational category, the framework explicitly addresses agentic AI through a dedicated sub-type: ‘unauthorised actions by autonomous AI agents’. This sub-type flags the risk of agents autonomously acquiring system access beyond their authorised scope or triggering unpredictable multi-agent chain reactions. Rather than imposing static risk tiers, it guides sectoral regulators through a four-step regulatory cycle: compiling an inventory of use cases, identifying risks against the 20 sub-types, assessing risks and implementing proportional responses.

In Singapore, the Monetary Authority of Singapore (MAS) published the industry-developed Safeguards for Agentic Finance at Runtime (SAFR) white paper, proposing runtime governance checkpoints to verify and record AI agents’ proposed actions before execution. This was accompanied by the Personal Data Protection Commission (PDPC) finalising its Advisory Guidelines on Use of Personal Data in Generative AI, clarifying data protection responsibilities across model training, fine-tuning and retrieval-augmented generation (RAG). The Infocomm Media Development Authority (IMDA) also issued voluntary Transparency Guidelines for Generative AI Chatbots.

Hong Kong’s financial regulators selected 36 use cases for the first Generative AI Sandbox++ cohort to test agentic applications, including testing how AI can oversee other AI applications. Separately, the Office of the Privacy Commissioner for Personal Data (PCPD) published dedicated guidance on agentic AI, warning that autonomous agents pose higher data privacy risks than standard chatbots and recommending a human-in-the-loop approach for decisions likely to have a significant impact on individuals, alongside minimum access controls.

Elsewhere in the region, New Zealand’s Financial Markets Authority launched a thematic review of AI in financial advice, while Standards New Zealand consulted on adopting 14 international AI standards, including the joint International Organization for Standardization (ISO) and International Electrotechnical Commission (IEC) standard ISO/IEC 38507 on the governance implications of organisational use of AI.

2. Cyber resilience against AI-enabled threats

Regulators in Australia, Singapore, Taiwan and the Philippines issued warnings or mandated measures in response to concerns that frontier AI models could compress incident response windows by automating the identification and exploitation of software vulnerabilities. The Australian Prudential Regulation Authority (APRA) and ASIC highlighted how these capabilities shorten the time available to patch systems, calling on boards to establish clear escalation pathways, timely patching schedules, tested backups and recovery arrangements before a crisis occurs. Taiwan’s Financial Supervisory Commission (FSC) highlighted ‘Claude Mythos Preview’ to illustrate frontier models’ ability to identify vulnerabilities and generate attack code at scale, warning of shorter remediation windows and a ‘patch flood’ that could overwhelm institutional information technology (IT) resources. To counter these threats, in the Philippines, the BSP issued six recommendations, urging financial institutions to maintain accurate, up-to-date inventories of externally reachable assets and dependencies and to transition to phishing-resistant authentication, such as hardware security keys, for all administrative and privileged access.

While defensive AI can support vulnerability detection and incident response, regulators cautioned that these tools require governance and human oversight and cannot replace fundamental cyber hygiene. Regulators also emphasised the systemic risks associated with shared dependencies, where concentrated reliance on a few cloud, software as a service (SaaS) or AI model providers could escalate an isolated vendor disruption into sector-wide failure. For software-supply-chain risks, Taiwan’s FSC recommended software bills of materials (SBOMs) to track third-party components, alongside contractually agreed patching service levels with critical suppliers.

In Singapore, MAS required key financial institutions to conduct AI-assisted red teaming on critical, internet-facing systems from 1 July 2026. MAS also announced the formation of the Association of Banks in Singapore’s AI-Driven Cyber and Tech Risk Taskforce (ABS-ACT), comprising MAS and industry representatives. Additionally, MAS announced that it will soon issue supervisory expectations requiring key institutions to submit detailed assessments and action plans regarding their vulnerability detection, patching, testing, and recovery capabilities.

3. Preparing for quantum-related security risks

Measuring and building resilience against future quantum computers capable of breaking widely used public-key cryptography is an ongoing concern for regulators. The Hong Kong Monetary Authority (HKMA) launched its Whitepaper on Quantum Preparedness of Hong Kong’s Banking Sector alongside its first Quantum Preparedness Index (QPI), which evaluates institutions across four dimensions: Awareness, Planning, Pilots and Practical Preparedness. The banking sector’s initial baseline score was 2.3 on a ten-point scale. Although 68% of surveyed banks were at the awareness, planning or piloting stages, 32% had not yet begun their transition journey. Around half lacked formal post-quantum planning, only half had discussed the risk at board level, and one-third were exploring or piloting quantum-related technologies. HKMA Deputy Chief Executive Arthur Yuen warned that banks must transition to post-quantum cryptography (PQC) before the first active decryption attacks occur, noting that the HKMA will monitor individual bank QPI scores.

To support its goal of achieving full sectoral readiness (a QPI score of 10) by 2030, the HKMA is co-developing a PQC toolkit with the School of Business and Management of the Hong Kong University of Science and Technology (HKUST). The toolkit aims to build ‘cryptographic agility’, a capability recommended in the white paper that enables banks to upgrade cryptographic algorithms without major system disruption.

In Singapore, MAS stated its aim for financial institutions to achieve quantum resilience before the end of the decade. MAS announced that it would issue supervisory expectations later in 2026, with progressive milestones and timelines for institutions to compile inventories of cryptographic assets, prioritise the migration of vulnerable systems, and build technical capabilities and governance. MAS also committed SGD 220 million over three years under the renewed Financial Sector Technology and Innovation (FSTI 4.0) scheme. The scheme has six tracks, with quantum technology, AI and distributed ledger technology (DLT) among the priorities supported through its Institution Project track.

4. Stronger authentication and fraud prevention

As generative AI increases risks from digital impersonation, supervisors strengthened authentication and fraud-prevention expectations. In Hong Kong, the Securities and Futures Commission (SFC) issued a circular expecting licensed internet brokers and SFC-licensed virtual asset service providers (VASPs) to implement phishing-resistant authentication, such as passkeys or robust device binding, for client logins and device registration or binding. The SFC stated that one-time passwords (OTPs) are not phishing-resistant and should not be used for these processes. Clients should generally register no more than three passkeys and/or three devices, with additional registrations permitted following adequate assessment. Firms must also monitor abnormal trading patterns, particularly following device changes or password resets, and report hacking incidents immediately. Robust authentication is expected as soon as practicable and no later than 8 July 2027, with immediate implementation by large internet brokers. Monitoring and response enhancements are expected immediately. The SFC will hold firms accountable for client losses where safeguards fail to prevent, detect and stop large-scale unauthorised transactions following hacking incidents.

Japan’s JFSA updated its supervisory guidelines for insurers, moneylenders and payment firms to make phishing-resistant multifactor authentication (MFA), such as passkeys or certificate-based logins, the mandatory default for logging in, withdrawing funds or changing payee accounts. Where customers cannot use the necessary equipment, firms must provide alternative MFA and monitor the use of exceptions. To prevent credential harvesting, the JFSA also expects firms to remove login links from direct customer communications, such as emails and short message service (SMS) messages, except where no alternative is available, for example to fulfil a legal obligation.

Onboarding and biometric systems have also become key defensive frontiers. The HKMA updated its remote onboarding guidance for banks and stored value facility (SVF) licensees in response to AI-driven automation and deepfake attacks. It reaffirmed two existing core principles: Identity Authentication, using technology to verify electronic document reliability, and Identity Matching, using technology to link the customer conclusively to the identity provided. In Thailand, the Bank of Thailand (BOT) consulted on security guidelines for facial recognition used in mobile banking apps, targeting AI deepfakes, presentation attacks and video-injection attacks. The draft guidelines propose at least quarterly vulnerability assessments, annual penetration testing covering injection attacks, and performance metrics including a maximum False Match Rate (FMR) of 0.01% and a maximum Impostor Attack Presentation Accept Rate (IAPAR) of 4% for each attack type.

To support these technical controls, jurisdictions are reinforcing legal deterrents and cross-sector coordination. Taiwan’s FSC amended its fraud-prevention regulations to enable cross-sector fraud-account inquiries involving deposit-taking institutions, payment firms and VASPs, and support shared anti-fraud platforms subject to approval. Malaysia’s Parliament passed the Cybercrimes Bill 2026, including a synthetic-media offence in Clause 23 that would criminalise distributing deceptive computer-generated or manipulated audio or visual content with intent to commit or facilitate an offence, carrying a penalty of up to seven years' imprisonment or MYR 500,000 in fines.

5. Digital asset regulation and tokenisation

Regional regulatory developments on digital assets largely focused on supervision, financial crime controls and tokenised settlement infrastructure.

VASP licensing and stablecoin frameworks

Taiwan promulgated the Virtual Asset Service Act, establishing a comprehensive regulatory regime for seven activities, including exchange, custody, transfer and lending of virtual assets. Stablecoin issuance will require approval from the Central Bank and FSC, full reserve assets held in trust and public disclosures. The law provides for penalties of three to ten years’ imprisonment for specified fraud and market-manipulation offences. Existing registered service providers will have 12 months from commencement to apply for a licence.

In Singapore, MAS consulted on amendments to the Payment Services Act 2019 to operationalise its Single-Currency Stablecoin (MAS-SCS) framework. Under the proposed framework, licensed Singapore-incorporated issuers may use the ‘MAS-regulated stablecoin’ label. The proposals would prohibit payment of interest to holders, require full reserve backing with segregated assets, and introduce regular stress testing. They also outline criteria for recognising a limited number of foreign-issued stablecoins under equivalent regulatory regimes and would permit jointly issued stablecoins subject to case-by-case exemptions and safeguards.

Wholesale tokenisation and settlement infrastructure

Concurrently, jurisdictions are using tokenisation and wholesale digital currencies to modernise core settlement layers. In Australia, the Reserve Bank of Australia (RBA) launched a consultation on the role of the Reserve Bank Information and Transfer System (RITS) in supporting a tokenised financial ecosystem, building on Project Acacia’s findings that tokenisation has the potential to improve wholesale market efficiency. Alongside this, the RBA and Treasury reaffirmed that there is currently no clear public interest case for a retail central bank digital currency (CBDC).

In the Philippines, the BSP highlighted findings from its Project Agila report detailing its wholesale CBDC proof of concept on Hyperledger Fabric. The pilot demonstrated round-the-clock transfer capability. In scalability tests, it processed 105,000 transactions in under 10 hours; further optimisation was needed at higher volumes. The report recommended focusing next on settling tokenised financial securities and institutional cross-border payments.

In Hong Kong, following the June 2026 completion of the first phase of the DLT Fixed Income Review by the Financial Services and the Treasury Bureau (FSTB) and the HKMA, clarifications confirmed that debenture holders’ registers maintained on DLT can meet the relevant Companies Ordinance record-keeping requirements. The Chief Executive’s Policy Address outlined plans to test tokenised Exchange Fund Bills and promote trading of regulated stablecoins on licensed virtual asset trading platforms.

Financial crime controls and retail safeguards

To support this expanding digital ecosystem, regional regulators are tightening financial crime prevention rules and retail market safeguards. Thailand’s Securities and Exchange Commission (SEC) proposed restricting stablecoin transfers to verified customer-owned accounts or wallets, with separate inbound and outbound limits of THB 5 million per customer per day per operator, subject to specified exemptions. This was accompanied by the formal issuance of Thailand’s Travel Rule for Digital Assets, effective 27 February 2027. Draft rules for crypto exchange-traded funds (ETFs) would initially cover Bitcoin and Ethereum and primarily require Thai-licensed custodians. A related consultation proposed revised qualifications for foreign custodians serving digital asset funds, including a possible list of 11 jurisdictions.

In Vietnam, the National Assembly passed a law which will bring crypto-asset service providers into the AML regime from 1 December 2026. The law subjects providers to customer due diligence and record-keeping, and introduces 15 crypto-specific suspicious activity indicators, such as transaction splitting and the use of mixers. It also places existing restrictions on foreign-currency transactions, pricing, listings and quotations on a statutory basis, subject to authorised exceptions.

We invite you to read the full report for the detailed updates by jurisdiction, with other notable regional developments included in the appendix.

Did you find this useful?

Thanks for your feedback