PYMNTS AI
Big Tech’s AI Gatekeepers Could Draw Regulatory Scrutiny
Legal experts highlight emerging regulatory scrutiny on Big Tech gatekeepers and liability attribution for autonomous AI agents.
Open sourceRegulatory intelligence13 jurisdictions26 primary instrumentsReviewed 21 Aug 2026
Global financial AI regulation
See where rules are binding—and what your control framework must do next.
Compare legislation, supervisory expectations and policy development across financial centres. Every position is dated, classified by legal force and linked to the regulator’s own evidence.
Regulatory change desk
What changed recently
Recent source-linked developments from the live OneBench corpus. The jurisdiction tracker below remains the authoritative statement of current position.
arXiv cs.LG — Machine Learning
Towards Reversible Forgetting: Managing Obsolete Knowledge in Continual Enterprise AI Agents
ArXiv paper proposes a framework for enterprise AI agents to selectively unlearn obsolete policies, regulations, and workflow rules.
Open sourceJurisdiction tracker
Compare the regulatory position
Filter by geography, legal force, implementation stage or control topic. Open a profile for the factual position, practical implications and primary instruments.
Compare regulatory systems
14 of 14 shownConsultationConsultation
Global standard setters
Global bodies are converging on board accountability, lifecycle controls, third-party concentration and system-wide monitoring, but do not create directly binding firm obligations.
Next milestone
Open regulatory profile FSB consultation closed on 22 July 2026; monitor publication of the final sound practices.
Phased implementationBinding law
European Union
The EU combines horizontal AI legislation with existing financial-services rules. Creditworthiness and access-to-service uses can be high-risk, while DORA governs operational resilience and critical technology dependencies.
Next milestone
Open regulatory profile High-risk Annex III requirements are scheduled for 2 December 2027 following the 2026 AI Omnibus changes.
In forceSupervisory expectation
United Kingdom
UK financial regulators currently apply technology-neutral rules rather than a new AI rulebook. Consumer Duty, SM&CR, governance and model-risk expectations remain the main control anchors.
Next milestone
Open regulatory profile Monitor the FCA’s promised publication of good and poor practice from its 2026 engagement and testing work.
In forceSupervisory expectation
United States
US oversight remains sectoral and use-case based. Banking agencies revised model-risk guidance in 2026, while fair-lending, adverse-action, privacy and consumer-protection laws continue to apply to algorithmic decisions.
Next milestone
Open regulatory profile Monitor the planned interagency request for information on AI, including generative and agentic systems.
Finalised — not yet effectiveSupervisory expectation
Canada
OSFI has finalised an enterprise-wide, risk-based model-risk guideline that expressly includes AI and machine-learning models across federally regulated financial institutions.
Next milestone
Open regulatory profile Guideline E-23 takes effect on 1 May 2027.
In forceSupervisory expectation
Australia
Australia is applying existing prudential and conduct obligations while supervisors press firms to close a widening gap between AI adoption and governance maturity.
Next milestone
Open regulatory profile Regulated entities should expect supervisory follow-up on the weaknesses APRA identified in its 2026 targeted review.
ConsultationConsultation
Singapore
MAS is moving from principles and industry toolkits towards detailed, proportionate AI risk-management expectations for all financial institutions, including GenAI and agents.
Next milestone
Open regulatory profile Consultation closed on 31 January 2026; final guidelines and the proposed transition period remain to be confirmed.
In forceSupervisory expectation
Hong Kong
HKMA combines current high-level AI and consumer-protection principles with supervised experimentation through an expanded GenAI sandbox.
Next milestone
Open regulatory profile Sandbox++ launched in March 2026; monitor resulting good practice and any follow-on supervisory circulars.
Policy developmentResearch & monitoring
Japan
Japan’s FSA is taking a technology-neutral, dialogue-led approach that recognises both AI risk and the ‘risk of not taking action’ in financial services.
Next milestone
Open regulatory profile Version 1.1 was published in March 2026; monitor policy clarifications arising from the public-private forum.
Policy developmentGuidance
India
RBI’s FREE-AI framework sets a financial-sector direction for responsible and ethical adoption, intended to balance innovation, inclusion and risk management.
Next milestone
Open regulatory profile Monitor the conversion of FREE-AI recommendations into formal directions and sector-specific supervisory expectations.
In forceSupervisory expectation
Switzerland
FINMA uses technology-neutral supervision and expects institutions to identify, limit and monitor AI risk through governance and risk-management systems proportionate to materiality.
Next milestone
Open regulatory profile FINMA continues risk-based supervisory exchanges and asks institutions to engage early on critical AI processes or regulatory-parameter use.
In forceGuidance
United Arab Emirates
The UAE now combines cross-authority enabling-technology guidance with a 2026 CBUAE consumer-protection note specific to responsible AI and machine learning in licensed financial institutions.
Next milestone
Open regulatory profile Monitor supervisory implementation of the February 2026 consumer-protection guidance and any sector-specific follow-on rules.
Policy developmentResearch & monitoring
South Africa
South African financial supervisors are developing a joint regulatory approach after surveying adoption and finding weaknesses in fairness testing, third-party model risk and board oversight.
Next milestone
Open regulatory profile Prudential Authority and FSCA are preparing a discussion paper setting out their regulatory approach.
Policy developmentResearch & monitoring
Brazil
Brazil’s central bank is studying AI use, risks and impacts across regulated institutions as part of its 2025–2026 regulatory priorities; no AI-specific financial rule is yet represented here.
Next milestone
Open regulatory profile Monitor publication of the central bank’s policy conclusions and any resulting consultation or rulemaking.
Control convergence
Where supervisors are focusing
An explicit-coverage map—not a score. It shows which control topics appear in the regulatory position tracked for each jurisdiction.
| Jurisdiction | Governance & accountability | Model risk & validation | Customer & conduct | Fairness & explainability | Data & privacy | Third party & resilience | Monitoring & reporting | GenAI & agents |
|---|---|---|---|---|---|---|---|---|
| EU | Explicitly covered |
Implementation horizon
Dated milestones to plan for
Only milestones with a confirmed date in primary-source material appear here. Undated policy intentions remain visible in the tracker without a fabricated deadline.
Canada
View profile Guideline E-23 takes effect on 1 May 2027.
European Union
View profile High-risk Annex III requirements are scheduled for 2 December 2027 following the 2026 AI Omnibus changes.
Evidence standard
Regulatory intelligence, not legal advice. OneBench records what the official source says, when it was reviewed and whether it is law, supervisory expectation, guidance, consultation or monitoring.
Regulator, central-bank and official legislative material is the evidence base.
Legal force and implementation stage are classified separately.
Practical implications are labelled and never replace the source position.