The SAFR (Safeguards for Agentic Finance at Runtime) white paper closes with an invitation: "FinTechs and financial institutions are invited to contribute to SAFR and the BuildFin.ai working group by sharing pilot findings, identifying gaps in the specification, and raising domain requirements not yet addressed." The invitation is real, and it rewards preparation: a contribution grounded in a governed pilot run gives the work group something it can examine. This guide covers who the invitation addresses, what the three contribution tracks mean in practice, how the Expression of Interest mechanism works, and the four artifacts to have in hand before you write anything.
Who the SAFR invitation is for
Safeguards for Agentic Finance at Runtime (SAFR) is a white paper, version 1.0, published in July 2026 under MAS's BuildFin.ai initiative. MAS established a dedicated work stream under BuildFin.ai to develop implementation resources for agentic AI risks, and the paper was written with eight industry members: Ant International, Circle, HSBC, J.P. Morgan Chase, Manulife, Mastercard, OCBC, and Visa.
The invitation in the paper's conclusion is addressed to FinTechs and financial institutions. In practice that means teams deploying AI agents in the applications the paper names as raising heightened governance concerns: payments, liquidity management, compliance triage, credit assessment, and post-transaction processing. If you are running or planning an agent in one of those areas, you are the intended contributor.
Before committing internal resources, note SAFR's status: it "does not constitute regulatory guidance or supervisory expectations." It is an industry reference approach for a runtime governance layer, and each institution remains responsible for determining how its deployment aligns with applicable supervisory expectations. Contributing is voluntary industry participation, and a chance to shape the specification while it is still forming. For a full walkthrough of the framework itself, see SAFR, Explained.
What "contributing" means, per the paper
The invitation names three tracks.
Pilot findings. Evidence from operating the reference approach: an agent bound to a mandate, its proposed actions evaluated pre-execution, the four dispositions (Deny, Escalate, Auto-Execute, Observe) producing a full audit record. The paper's own operational sections signal what findings matter: escalation volume (an escalation function that generates more reviews than the institution can process defeats its own purpose), review turnaround (timeout windows that reflect realistic reviewer availability), and reviewer authority. Measured data on those three dimensions, from real traffic, is exactly what a version 1.0 specification lacks.
Gaps in the specification. Places where the four components, the Governance Envelope, or the mandate model meet a scenario the paper does not resolve. Institutions operating in open networks, where an agent may be registered in multiple identity databases, are one likely source of edge cases the closed-loop pattern does not surface.
Domain requirements not yet addressed. Needs specific to a business line, asset class, or jurisdiction that the current control categories and calibration factors (action reversibility, financial materiality, customer impact severity, regulatory sensitivity, novelty) do not cover. A domain where those five are insufficient is a contribution in itself.
How BuildFin.ai submissions work
The MAS media release accompanying SAFR, published 3 July 2026, states that "Interested industry partners are invited to join the BuildFin.ai work group to contribute to and help shape subsequent iterations of SAFR" and links an Expression of Interest form as the submission channel.
The form is a Singapore Government FormSG form titled "BuildFin.ai: Expression of Interest" at form.gov.sg/692d54b9d4e0e288131feb1a, also reachable via go.gov.sg/buildfin-registration. Fields include organization name, country of incorporation, and contact person. Submissions are evaluated by MAS, and the form notes that "discretion and caution should be exercised in the provision of proprietary or confidential information." No deadline is stated, and no named contact or email is published; the form is the mechanism.
BuildFin.ai works with financial institutions developing AI capabilities, technology providers with relevant AI solutions, and research institutions contributing to financial AI development. If your organization fits one of those profiles and operates or plans an agent in the heightened-concern applications, the form is where a pilot contribution starts. Per the same release, the Future of Finance Institute (FFI), announced 25 June 2026, "will support future adoption of the SAFR framework through the facilitation of industry pilots and sandbox experimentation."
What a credible pilot contribution contains
The contribution track with the highest bar and the highest value is pilot findings. Before drafting a submission, assemble four artifacts. Each maps to a section of our SAFR Readiness Checklist, so you can pressure-test the pilot design before running it.
A submission with these four artifacts lets the work group examine a working instance of the specification. Gaps and domain requirements land harder when a logged governance decision illustrates them.
| Artifact | What it demonstrates | Checklist section |
|---|---|---|
| A scoped agent workflow | One agent, registered with a named owner, doing one job from the heightened-concern list (payments, liquidity management, compliance triage, credit assessment, post-transaction processing) | A. Agent inventory & identity |
| A mandate definition | Delegated authority in explicit, machine-readable form: what the agent may do, within what limits, under what conditions | B. Mandates & authorization |
| Disposition calibration | All four outcomes configured, with thresholds tied to the paper's calibration factors and a record of how tuning moved actions between Auto-Execute and Escalate | D. Disposition engine; F. Escalation operations |
| Evidence samples | Audit log entries capturing what SAFR requires of each: the envelope as submitted, the mandate checked, the outcome, the specific rules applied, the basis for the outcome, and time elapsed at each stage | G. Audit & evidence |
SAFR pilot timeline realism
Producing those four artifacts is a multi-week effort, dominated by elapsed time you cannot compress. Disposition calibration needs governed traffic to tune against. Escalation turnaround can only be measured by routing real Decision Requests to real reviewers across real coverage windows, overnight and weekend included.
A workable shape is four weeks: scope the workflow and author the controls in week one, run governed traffic in simulation and calibrate dispositions in week two, operate live escalations and tune thresholds in week three, export and independently verify the evidence in week four. The full week-by-week plan is on our SAFR implementation page. No submission deadline is stated on the BuildFin.ai Expression of Interest form as of July 2026. If BuildFin.ai publishes submission windows, work backward from them with at least that four-week runway.
Key Takeaways
The invitation asks for evidence from practice, and the strongest contributions will come from institutions that arrive with a governed run and its records. That standard is reachable within a quarter for a single scoped workflow. KLA Control Plane implements the SAFR pattern, shipping today: Agent Registry, Policy Builder, KLA Policy Engine, and an append-only Audit Trail that exports Sealed Evidence Bundles verifiable offline. KLA is the independent runtime layer for SAFR-aligned pilots — talk to us before you submit. Start with the SAFR Readiness Checklist to find your gaps, then book a 30-minute SAFR gap review. A thirty-minute session covers one governed run end to end: policy, decision, escalation, sealed evidence. Source: Safeguards for Agentic Finance at Runtime, white paper v1.0, MAS BuildFin.ai, July 2026. Quoted passages © Monetary Authority of Singapore. SAFR is an industry reference and does not constitute regulatory guidance. KLA is independent of and not affiliated with, endorsed by, or certified by MAS or BuildFin.ai.