AI Regulation Tracker / Regulator guidance
Taiwan's FSC AI Guidelines put algorithmic accountability on banks, insurers and securities firms
Taiwan's Financial Supervisory Commission has set six core principles and eight supporting policies for every financial firm using AI. The guidance is non-binding, but sector self-regulation turns it into day-to-day duty.
Taiwan's Financial Supervisory Commission has drawn a single line around the whole financial sector's use of AI. On June 20, 2024, the FSC finalized its Guidelines for the Application of AI in the Financial Industry, the capstone of a project it began on October 17, 2023, when it published six core principles and a set of supporting policies. The result folds banks, insurers, and securities and futures firms under one supervisory umbrella rather than issuing separate rulebooks for each.
The guidelines are not a statute and carry no direct penalties. They are administrative guidance, soft law in the FSC's own framing. What gives them force is the mechanism the FSC built into them: one of the eight supporting policies directs industry associations to write self-regulatory standards that translate the principles into concrete operating rules. In practice, the principles set the ceiling and the association rules set the floor a firm is measured against.
The six principles
The FSC's framework rests on six core principles: governance and accountability; fairness and human-centric design; privacy and customer rights; robustness and security; transparency and explainability; and sustainable development. Read together, they describe an AI system that a named part of the organization owns, that is tested against bias, that protects customer data, that resists failure, that can explain its outputs, and that is deployed with a view to long-term stability rather than short-term gain.
Governance and accountability is the load-bearing principle for compliance officers. It pushes responsibility for an AI system up to a level of the firm that can answer for it, rather than leaving it with the team that built or bought the model. Fairness and human-centric design is where algorithmic bias mitigation lives. Transparency and explainability is what a firm leans on when a customer or a supervisor asks why a model reached a particular result.
What firms are expected to do
The operational expectation is risk-based governance across the full AI system lifecycle, not a one-time review at launch. A bank or insurer using AI is expected to identify and reduce algorithmic bias, keep model decisions explainable, and manage third-party and vendor AI risk through contract terms rather than trust. Where a firm relies on an outside AI vendor, the guidelines expect the risk to be allocated and monitored in the agreement, so accountability does not disappear at the boundary between the firm and its supplier.
Because the guidelines run sector-wide, the same principles reach a securities firm's trading models, an insurer's underwriting and claims tools, and a bank's customer-service systems. The self-regulatory standards written by each industry association are where the sector-specific detail is filled in.
What the guidelines do not do
The guidelines do not create new fines, license conditions, or a private right of action, and they do not by themselves make any single AI use unlawful. They are guidance, and a firm that ignores them is not automatically in breach of a statute. The exposure runs through the associated self-regulatory standards and through the FSC's existing supervisory relationship with regulated firms, not through the guidelines as a freestanding penalty regime. Reading them as binding law overstates their legal weight.
The cross-border read
For a US firm, the closest reference point is model risk management under the Federal Reserve and OCC's SR 11-7, which pushes governance, validation, and documentation onto banks that rely on models. Europe reaches similar ground through DORA's operational-resilience duties and the AI Act's risk tiers. Taiwan's distinctive move is consolidation: instead of layering AI duties onto separate banking, insurance, and securities regimes, the FSC set one principle-based standard for all financial AI and let each industry association operationalize it. A US or European firm operating in Taiwan will not find binding penalties in the guidelines themselves, but should expect the relevant association's self-regulatory standard to define what compliant AI governance looks like in practice.
Frequently Asked Questions
What did Taiwan's FSC actually change?
The FSC finalized its Guidelines for the Application of AI in the Financial Industry on June 20, 2024, setting six core principles and eight supporting policies for financial firms using AI, and directing industry associations to write self-regulatory standards that put them into practice. The guidelines build on principles the FSC first published on October 17, 2023.
Who is affected?
Banks, life and non-life insurers, and securities or futures firms in Taiwan that use AI, across customer-facing functions like product recommendation and service and operational functions like underwriting, claims, and risk modeling.
Are the guidelines legally binding?
No. The FSC frames them as non-binding administrative guidance, or soft law. They carry no direct penalties. Their practical force comes from the sector self-regulatory standards that industry associations write to operationalize them, together with the FSC's existing supervision of regulated firms.
What should a compliance team do first?
Inventory every AI system in production, assign clear governance ownership for each, and test each against the six principles, with particular attention to bias mitigation, explainability, and whether vendor contracts allocate AI risk. Then confirm which association self-regulatory standard applies.
Sponsored Training
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.