SAFR (Safeguards for Agentic Finance at Runtime) is a reference approach for governing AI agents in financial services at the moment they propose to act. It was published in July 2026 as a whitepaper under BuildFin.ai, developed with contributions from financial institutions and payment networks.

Start here

  • What SAFR is: the framework in plain language, then in detail.
  • What SAFR is not: where it sits relative to regulation, payment rails, compliance platforms, and LLM guardrails.
  • Glossary: every term, labelled by whether it comes from the whitepaper or from OpenSAFR.
SAFR is an industry reference approach. The whitepaper states explicitly that it is not regulatory guidance or a supervisory expectation. OpenSAFR is an independent, unofficial implementation.

The four components

SAFR whitepaperp. 9

Agent Identity

The component that binds a proposed action to a verified, registered agent before any other evaluation.

SAFR whitepaperpp. 9–10

Controls Repository

The institution's configurable rulebook of controls drawn from policy, regulation, product rules, and mandates.

SAFR whitepaperpp. 10–11

Disposition Engine

The component that evaluates an action against applicable controls and resolves it to one of four outcomes.

SAFR whitepaperpp. 11–12

Audit Log

A tamper-evident, append-only record of every governance decision.