What Is a FINMA‑Aligned Technical Blueprint for Agentic AI?

How Switzerland Is Turning AI Governance Principles into Code

Key Takeaways

• A FINMA‑aligned technical blueprint for agentic AI turns high‑level supervisory principles into testable technical controls for AI systems used in banking.
• LatticeFlow AI and Unique AI created the first such blueprint, assessing a live Investment Insights Agent used to generate personalised investment rationales for relationship managers.​
• The blueprint maps FINMA guidance 08/2024 onto concrete checks for robustness, monitoring, explainability and human‑override in real banking conditions.​
• It gives banks audit‑ready evidence that AI behaves reliably, is understandable to human users and can be overridden, supporting prudential and conduct‑risk expectations.
• This Swiss initiative offers a template for AI governance in financial services that other regulators and global banks can adapt.

Agentic AI in Banking

Agentic AI in banking refers to AI systems that can autonomously perform multi‑step tasks—such as drafting investment rationales or assisting with KYC—while interacting with users and other systems. These agents support, but do not replace, human decision‑makers, which makes governance, monitoring and override capabilities critical in regulated environments.

FINMA‑Aligned Technical Blueprint

A FINMA‑aligned technical blueprint is a framework that translates FINMA’s AI governance guidance into concrete, measurable technical controls and tests for AI systems in financial services. It helps banks demonstrate that AI behaves reliably under real‑world conditions, is transparent enough for scrutiny and remains under effective human control.

What Is a FINMA‑Aligned Technical Blueprint for Agentic AI?

In February 2026, LatticeFlow AI and Unique AI announced what they describe as the first FINMA‑aligned technical blueprint for governing agentic AI systems in Swiss financial services. The blueprint applies FINMA’s 2024 guidance on AI to a live Investment Insights Agent, an AI solution that helps relationship managers and client advisers generate personalised investment recommendations and rationales faster and more consistently.

Rather than relying on static documentation or one‑off model approvals, the blueprint evaluates how the agent behaves in production under real banking conditions—how it responds to changing inputs, interacts with users and maintains reliability over time. The goal is to move from abstract principles to decision‑ready evidence for risk, compliance and audit teams.​

How the Blueprint Maps FINMA Principles into Technical Controls

According to LatticeFlow, the blueprint maps FINMA guidance 08/2024 to a set of technical control dimensions, including:​

• Robustness and reliability: Testing how the agentic AI system behaves across a range of realistic scenarios, including noisy inputs and edge cases.
• Monitoring and observability: Ensuring ongoing monitoring captures performance drifts, anomalies and potential failure modes in a banking context.​
• Explainability and transparency: Evaluating whether outputs can be understood, challenged and documented by human users and oversight functions.
• Human‑in‑the‑loop and override: Assessing whether relationship managers and advisers can effectively override or correct AI recommendations when needed.

By grounding these dimensions in test cases and metrics, the blueprint offers a playbook for turning supervisory expectations into code‑level checks that can be audited.

Why Agentic AI Needs This Level of Governance

Agentic AI systems in banking are not just chatbots; they can shape investment ideas, client conversations and documentation at scale. That makes them part of the bank’s fiduciary and conduct‑risk footprint, not just a UX feature.​

The ECB has warned that AI “does not dilute responsibility” and must be governed as a core risk topic, embedded in risk appetite, pre‑implementation assessments and post‑implementation monitoring. The FINMA‑aligned blueprint is one way to show that responsibility in practice: it provides concrete evidence on whether the AI behaves consistently, supports sound decisions and stays within defined boundaries.​

A Simple Framework for Banks Using FINMA‑Style AI Blueprints

1. Select high‑impact AI use cases
   Focus first on agentic AI systems that influence investment advice, KYC decisions, AML alerts or other high‑stakes outcomes.
2. Map regulatory guidance to technical tests
   Translate supervisory expectations (for example, FINMA or ECB principles) into testable requirements across robustness, monitoring, explainability and human‑override dimensions.
3. Run blueprint assessments in production‑like conditions
   Evaluate how the AI behaves using real or representative data, multi‑step workflows and live user interactions, not just lab benchmarks.​
4. Generate audit‑ready evidence and dashboards
   Produce artefacts that risk, compliance and internal audit teams can review—test results, metrics and narratives that demonstrate control and traceability.
5. Continuously monitor and iterate
   Treat the blueprint as a living framework: update tests and controls as models, data and regulatory expectations evolve, ensuring AI remains aligned with risk appetite and governance.

FAQ: FINMA‑Aligned Agentic AI Blueprint and Banking

What is the FINMA‑aligned technical blueprint for agentic AI?
It is a framework created by LatticeFlow AI and Unique AI that maps FINMA’s AI governance guidance to concrete technical tests and controls for agentic AI systems used by banks.

Which AI system was assessed with this blueprint?
The blueprint was applied to Unique AI’s Investment Insights Agent, which helps relationship managers and client advisers generate personalised investment rationales and recommendations.

Why does this matter for fintech and banks?
It shows how to convert high‑level AI governance principles into audit‑ready evidence, helping institutions prove that AI is reliable, explainable and under human control in real‑world banking use cases.

Can the blueprint be used beyond Switzerland?
While it is aligned with FINMA guidance, the underlying approach—mapping principles to tests—is adaptable to other regulatory regimes and can inform AI governance for banks in different jurisdictions.

Sources

1. Technology is neutral, governance is not: AI adoption in the banking sector
2. Unique AI x LatticeFlow AI: Introducing FINMA-Aligned Technical Blueprint
3. Technology Is Neutral, Governance Is Not: What AI Governance in Banking Should Learn from Nonbank Systemic Risk
4. Opening Remarks for the “AI and Productivity across the Economy” Panel