Fidacy is an external, deterministic control with a tamper-evident audit and signed verdicts, the architecture regulators and risk teams ask for. Below is the dual EU/US mapping. These are alignment claims: Fidacy provides the evidence artifacts; your institution maps them to its own obligations.
The de-facto US framework for trustworthy AI, built on four core functions. Fidacy supplies concrete mechanisms and evidence under each.
GOVERNVersioned, activatable risk policies with every change recorded; an admin audit log of who changed what, when. Decisions reference the exact policy version that produced them.
MAPKnow-Your-Agent identity resolution by RFC 7638 thumbprint plus decision classification, each action is mapped to an agent, a mandate, and a risk context before it is scored.
MEASUREA deterministic Trust Score with the exact signals behind it, backtesting against historical traffic, and per-decision metrics (rates, confidences, rejection reasons).
MANAGEFail-safe degradation to review on any fault, webhooks for human-in-the-loop handling, a signed verdict for every decision, and a hash-chained audit trail anchored to Bitcoin for after-the-fact review.
Where audit evidence gets adjudicated, control by control. Coverage is honest: full means evidenced end to end for gated actions, partial means evidenced with named limits, moderate means supporting evidence only.
{{ row.control }}
{{ row.req }}
{{ row.mech }}
{{ row.coverage }}Released Feb 19, 2026: 230 control objectives across 7 risk domains and 4 adoption stages, voluntary and industry-led, structurally aligned with NIST AI RMF. Fidacy does not certify against it, no one does. It provides the evidence artifacts an institution maps to its own control objectives: signed verdicts, the policy-version record, the tamper-evident audit chain, decision metrics, and backtests. Because the FS AI RMF aligns with NIST, the mapping above carries directly into a Treasury-framework control narrative.
SB 26-189 (signed May 14, 2026, effective Jan 1, 2027) regulates automated decision-making technology in "consequential decisions", explicitly including financial services, under a disclosure model. Fidacy supplies what a deployer needs: developer disclosure, testing documentation (backtests, decision metrics), audit cooperation via the re-verifiable trail, and indemnification terms in the enterprise agreement.
Fidacy is built for automated decisioning in a high-risk-adjacent context, where the law expects governance, record-keeping, and human oversight. The append-only, hash-chained audit trail, sealed with Merkle anchors and re-verifiable offline, is the non-repudiable evidence automated-decision systems are expected to keep. Fail-safe degradation to review and the webhook path give the human-oversight hooks the Act calls for.
Under the 2026 Digital Omnibus, Art. 50 chatbot transparency applies Aug 2, 2026; high-risk record-keeping (Annex III) was deferred to Dec 2, 2027. Fidacy is not the high-risk system, it is the external control and evidence layer for the economic actions an AI system takes.
{{ row.control }}
{{ row.req }}
{{ row.mech }}
{{ row.coverage }}The full mapping, including notes and gap narratives, is published machine-readable at fidacy.com/compliance/mappings.json, a GRC tool or an AI assistant can consume it directly.
Commission Implementing Decision (EU) 2026/2274 cited EN 18286:2026 in October 2026, the quality management system standard for AI Act purposes. A provider of a high-risk system that applies it is presumed to meet the QMS duties of Art. 17, for that system and only for the clauses listed in its Annex ZA. Fidacy is not that provider and does not claim conformity with EN 18286. What it supplies is evidence that the procedures the standard asks for actually ran, action by action.
Art. 17(1)(a)
Compliance strategy, including management of modifications
Each verdict carries the policy_version that produced it, and each action decision receipt issued since October 18, 2026 carries the commit of the Fidacy engine that decided, both inside the signature.
Art. 17(1)(h)
Post-market monitoring
The decision feed and the signed Control Coverage Report, which states what was gated, what was not, and how strongly each agent and machine was identified.
Art. 17(1)(i)
Reporting of serious incidents
A per-decision incident pack: the signed receipt, the anchor receipt and the audit position, exportable for the authority.
Art. 17(1)(k)
Record-keeping of relevant documentation
The append-only audit chain, Merkle-checkpointed and anchored to Bitcoin, and a signed export for any period within your plan's retention window.
For a deployer, this is the part of the contract that EN 18286 cannot give on its own: a certificate describes the process, these records show it ran on the system you bought.
Under the Digital Operational Resilience Act (Regulation (EU) 2022/2554, applicable since January 17, 2025), a bank, insurer or payment institution using Fidacy treats it as an ICT third-party service provider. The obligations are the financial entity's. This is what Fidacy brings to that assessment, stated only where the system shows it.
sales@fidacy.com.Read this honestly. These are alignment claims, not a certification and not legal advice. Coverage describes the evidence artifacts Fidacy supplies for the gated slice of your system; mapping them to your institution's obligations is your compliance team's call.
GDPR. Fidacy minimizes personal data on the decision path, the optional reasoning layer receives only non-sensitive signals, never raw PII or payment credentials. Residency, retention, and erasure live in Data Protection; a DPA is available on request.
One mechanism, many frameworks. The same primitives, signed verdicts, a versioned policy record, and a tamper-evident audit chain, are the evidence that satisfies the EU AI Act, maps to NIST and the Treasury FS AI RMF, and backs Colorado ADMT disclosures. You instrument once and answer to all of them.