FCA Financial Crime · SYSC 6.3 · Proceeds of Crime Act
Fraud Detection & FRAML
- Decision reasoning captured at action level — not just the outcome
- Model version pinned to every block or pass decision
- Human escalation pathway logged and timestamped
Your AI agents are making decisions you can't explain
Your AI agents are making decisions you can't explain
Your AI agents are making decisions you can't explain
Your AI agents are making decisions you can't explain
A complete, regulator-ready audit trail for every action your AI agents take — before the FCA asks for one.
Evidence readiness
0%
Agent registry
0/15
Policy-bound actions
0
Human reviews
0.0%
Live intent feed
Filters
View
Time
Agent
Intent summary
Policy
Oversight
Status
09:46:58
AML Transaction Monitor V2.1
Flagged transaction TXN-99201 for structuring pattern
MAR 1.2
Post-exec
Flagged
09:45:48
Fraud Detection Agent V2.0
Reviewed fraud model performance metrics for drift
FCA SM&CR
None
Compliant
09:37:05
Trade Reporting Agent V1.4
Submitted final EMIR trade reports for the day
MiFID II
Pre-exec
Approved
09:26:38
Retail Credit Decision Agent V2.3
Generated daily credit portfolio risk summary
GDPR Art. 22
Pending review
Pending
09:16:42
Customer Communication Agent V3.0
Compiled Article 20 data portability package
GDPR Art. 20
Blocked
Blocked
The standard, mapped to what matters:
FCA SM&CR
Accountability chain
EU AI Act
Article 17 compliance
GDPR 22
Decision rights
100%
Open standard
Three questions you cannot answer today
Three questions you cannot answer today
Regulated firms deploying AI agents are building on the same unstable ground: the decisions are happening, the accountability isn't captured.
The FCA called
Your AI agent made 40,000 credit decisions last quarter. You have usage logs — but no structured record of the reasoning, the model version, or which human reviewed edge cases. The FCA has given you three weeks...
Enterprise clients want evidence
Procurement asked whether your AI systems are EU AI Act compliant. Your legal team said yes. Now they want documentation — and producing it means weeks of forensic work across logs no auditor can read.
You upgraded the model
In January you ran one model version. In March you switched. Something changed in the outputs — and nobody can say exactly when, because model version isn't captured at the action level.
The industry standard for AI agent accountability
The industry standard for AI agent accountability
The open standard for AI agent audit trails. Model, framework, infrastructure — independent of all three. Built by Arbiris. Free to use. Free to adopt.
FIG. 1.1
Agent Identity
Every agent is registered before deployment with its owner, version, model, policy scope, and accountable SMF.
FIG. 1.2
Intent Declaration
Every material action is captured as a signed record of context, reasoning, outcome, policy authority, and oversight.
FIG. 1.3
Policy Binding
Every action is linked to the versioned policy, rule, or regulatory instrument that authorised it at execution.
FIG. 1.4
Human Oversight
Every action declares the oversight actually applied: pre-execution, post-execution, or none, under approved tier rules.
AARF Level 1Foundation
5/5 completeComplete
All mandatory intent record fields captured
Cryptographic signing implemented
Evidence pack generation (manual compilation acceptable)
Human oversight requirements met
7-year retention
AARF Level 2Operational
3/4 completeIn progress
Automated evidence pack generation (Section 7) — checked
Exception queue with defined SLA — checked
Consumer duty outcome field populated
Policy coverage review — overdue
4 action types have not been reviewed this quarter, including Retail Credit Decision Agent v2.3.
Accountable SMF: J. Chen · SM&CR-441
AARF Level 3Full compliance
0/4 completeNot started
EU AI Act Article 14 format evidence packs available
Counterfactual summary populated (GDPR Article 22 scope)
Accountable SMF declaration on evidence packs
Annual third-party AARF audit
FIG 1.5 - AARF FRAMEWORK
Open, versioned, free to adopt.
Most organisations deploying AI agents in 2026 have AI policies. What they lack is a structured specification for what should be captured and evidenced every time an agent takes a decision affecting a customer, transaction, or regulated outcome.
AARF defines the minimum viable audit trail for compliant agentic AI: intent records, cryptographic signatures, policy references, oversight status, retention, and evidence packs.
The framework lets teams adopt incrementally while showing exactly where they stand against a published standard, not an internal scorecard.
Four pillars
The floor below which no regulated firm should build. Free to adopt. Maintained by Arbiris. Mapped to FCA SM&CR, EU AI Act, GDPR Article 22, and FCA Consumer Duty.
From agent decision to regulator-ready record.
From agent decision to regulator-ready record.
Instrument every decision in three lines. Monitor in plain English. Generate regulator-ready evidence in one click. No architecture changes required.
FIG 2.1 – INSTRUMENT
Capture every material decision in three lines.
Drop the Arbiris SDK into your existing agent code. It wraps around agent actions and captures a structured, cryptographically timestamped record of everything material—automatically, without changing your architecture.
- LangChain, LlamaIndex, AutoGen, and raw API support
- Model version pinned to every individual action
- Tamper-evident proof receipt generated instantly
- Policy reference binding at instrumentation time
FIG 2.2 – MONITOR
A dashboard compliance officers can actually use.
Every agent action, in plain English, mapped to the regulatory framework you operate under. Exception queues for human review. Real-time policy coverage gaps. No engineering degree required.
- Real-time intent feed with risk-level triage
- Workflow explorer reconstructs multi-step agent chains
- Exception queue with one-click approve / escalate / reject
- Policy coverage gaps surfaced automatically
FIG 2.3 – EVIDENCE
Regulator-ready documentation. One click.
Select a data range and a regulatory framework. Arbiris generates a formatted, audit-ready document — structured to EU AI Act Annex IV requirements — that you can hand directly to a regulator or external auditor.
- EU AI Act Annex IV technical documentation format
- Full decision chain visible — every agent step reconstructed for review
- 10 years tamper-evident retention, built in from day one
- Cryptographic proof receipts for every action included
Every agent. Every obligation. Covered.
Every agent. Every obligation. Covered.
Every sector operates under different mandates. Arbiris maps your agent actions to the frameworks that govern them — out of the box.
JMLSG Guidance · Proceeds of Crime Act
AML & Transaction Monitoring
- Trigger event and threshold captured per transaction
- Reasoning chain retained for every alert — cleared or escalated
- Named Senior Manager accountable for monitoring programme
OFSI · UK Sanctions Regulations · UN Consolidated List
Sanctions Screening
- Match / no-match decision recorded with list version at time of screening
- False positive resolution documented with human reviewer identity
- Policy reference bound to each screening action
Consumer Duty · CONC · GDPR Article 22
Credit decisioning
- Affordability assessment inputs retained per application
- Counterfactual explanation available for every declined decision
- Consumer Duty outcome classification recorded at action level
EMIR · MiFID II · MAR 1.2 · FCA TR Q&As
Trade Reporting
- Report submission linked to the agent action that generated it
- Amendment and cancellation events logged with reasoning
- Market abuse indicator checks captured per trade
FCA KYC Guidance · GDPR Article 22 · MLR 2017
KYC & Onboarding
- Identity verification outcome and data sources retained per application
- Risk classification reasoning captured at onboarding point
- Human review checkpoint logged where enhanced due diligence applies
Built for global compliance
Built for global compliance
FCA, EU AI Act, UK AI Safety, and GDPR — one audit trail, one evidence pack, one platform without rebuilding your documentation process.
EU AI Act
Requirements
Robustness, Accuracy, and Cybersecurity with technical logging of all high-risk interactions.
Arbiris solution
Hardware-Rooted Proof of every agentic intent anchored in secure enclaves for immutable auditbility.
UK AI Safety
Requirements
Transparency and audit trails across every autonomous agentic deployment.
Arbiris solution
Sub-second Cryptographic Receipts for 10-year retention providing a tamper-proof history of autonomous reasoning.
FCA
Requirements
Fiduciary duty in autonomous advice via provable retail investor alignment.
Arbiris solution
Statutory Logic Enclaves that prevent unauthorized trades by enforcing real-time compliance with fiduciary mandates.
FIG 3.0 — DESIGN PARTNER PROGRAMME
Don't let liability kill your AI roadmap.
Don't let liability kill your AI roadmap.
Five design partners will shape Arbiris and the AARF framework from the ground up — early access, roadmap input, and locked-in pricing.
Early access
Full platform access from day one, before general release.
Co-development input
Monthly sessions to shape the roadmap around your compliance workflow.
Preferential terms
Locked-in pricing and priority onboarding for the first cohort.
Apply for design partner access
5 spots remaining · 2-minute application