Live AI Governance Evidence for UK Financial Services

What this demonstrates

The AI Control Plane (ACP) governs what AI agents are allowed to do — and proves it. The evidence on this page demonstrates that ACP can:

• block agent actions that lack a valid delegation mandate;
• enforce the scope and constraints of delegated authority;
• apply the same policy decision across every surface (UI, API, CLI, MCP);
• record each decision as signed, tamper-evident evidence with a named accountable owner;
• block a production release when proof evidence is stale or failing.

Everything here is produced by the controls themselves — a nightly automated proof run plus reviewed attestations. Where something is not yet verified, this page says so.

The 60-second view for compliance, risk, security and audit

• What ACP controls: which actions AI agents may take, under whose delegated authority, with the same policy decision enforced across UI, API, CLI and MCP.
• What has been tested:fail-closed mandate enforcement, live credential revocation, decision-time SM&CR attribution, tamper-evident decision evidence, and an automated release gate — via the nightly proof run and reviewed attestations listed on this page.
• What remains unverified or firm-owned:SOC 2 Type II and independent penetration-test evidence are planned, not held today; firm-specific governance mapping, production outcome metrics and compliance sign-off remain the regulated firm's responsibility.
• Why this evidence matters:SM&CR accountability and Consumer Duty reviews need decision-time evidence of who authorised what. ACP produces that evidence as the control executes, not reconstructed after the fact.

The live proof status above and the tested-capability list below give a first-time reviewer the current position without leaving this page.

Latest Proof-Pack Evidence

• Generated: 2026-07-20 07:32:14 UTC.
• Overall status: pass.
• Critical steps: 8 covered, 0 skipped.
• Evidence-chain check: pass.
• Manifest signature: verified (HMAC-SHA256).
• Run, commit and artifact identifiers are recorded in the signed manifest and are available for inspection under NDA as part of an evidence review.

Freshness: proof evidence is treated as stale once it is older than 48 hours (the documented freshness threshold); the release gate additionally enforces a 26-hour maximum evidence age at deploy time. Current evidence age: about 4 hours.

Assurance note: Keyless provenance attestation is currently unavailable for this run. The HMAC-signed manifest still covers evidence integrity; this is an assurance note, not a proof-run failure.

Retention: each run's HMAC-signed manifest artifact is retained for 90 days, and the run-summary history is preserved permanently under version control.

Architecture & Governance

How the AI Control Plane governs an AI action and turns it into tamper-evident, signed evidence that risk, compliance, and audit can review. These diagrams are a conceptual overview for external review and intentionally exclude implementation and deployment detail.

Capabilities

Tested

Each tested capability names its evidence source: an automated proof run, or a human-attested review with its evidence identifier, review owner and review dates.

• Delegation mandates enforced fail-closedAn agent action without a valid delegation mandate is denied, an action inside a mandate's scope is allowed, and an action outside the mandate's constraints is denied. Each outcome produces a signed decision record.Evidence: Signed governance decisions (mandate-required deny, in-scope allow, out-of-scope deny) captured in a governed sandbox run and retained in the tenant evidence chain.(human-attested — Evidence ID ACP-EV-020 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Reviewed July 2026 · Next review due January 2027 · Evidence available under NDA)
• Agent credential issue, rotation, and revocationAgent credentials can be issued, rotated, and revoked while an agent is active. A live revocation was exercised against an active credential and the revocation appeared in the audit trail.Evidence: Credential and mandate revocation audit events from a governed sandbox run, including a revoke executed while the agent was active.(human-attested — Evidence ID ACP-EV-021 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Reviewed July 2026 · Next review due January 2027 · Evidence available under NDA)
• Decision-time SM&CR attributionGovernance decisions record the accountable senior manager at decision time, so each decision carries its accountability owner rather than reconstructing it afterwards.Evidence: Nightly foundation-proof evidence_chain step: SM&CR owner is set, a governed execution runs, and the decision evidence is retrieved with attribution present.(automated proof run)
• Policy parity across UI, API, CLI, and MCPThe same policy decision applies whichever surface initiates the action; there is no route that bypasses policy evaluation.Evidence: Cross-surface policy checks exercised across the management UI, tenant API, CLI, and MCP server against the same tenant policy.(human-attested — Evidence ID ACP-EV-022 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Reviewed July 2026 · Next review due January 2027 · Evidence available under NDA)
• Agent activity timeline evidenceEach governed agent has a retrievable activity timeline linking its actions, decisions, and audit events for review.Evidence: Nightly foundation-proof write_audit_trail step: audit rows and per-agent activity are retrieved and asserted non-empty.(automated proof run)
• Governance decision evidence with integrity chainEach governance decision produces a signed evidence record with parent-hash and payload-hash links, so tampering with the sequence is detectable.Evidence: Nightly foundation-proof evidence_chain step: decision evidence is retrieved and its hash chain and signature fields are verified.(automated proof run)
• Credential and API-key log redactionCredential material and API keys are redacted from logs and audit output, so evidence can be shared without exposing secrets.Evidence: Redaction checks over audit and log output during governed runs; no credential material present in retained evidence.(human-attested — Evidence ID ACP-EV-023 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Reviewed July 2026 · Next review due January 2027 · Evidence available under NDA)
• Automated proof pack and release-gate controlsA nightly automated run exercises the critical governance steps end-to-end and signs a manifest; deployment is gated on a recent passing manifest.Evidence: Proof Pack — Nightly workflow manifest (HMAC-signed) and the CD release gate that verifies it before deploy.(automated proof run)

Planned and firm-owned

Items below are not tested ACP capabilities: planned items are on the roadmap, and firm-owned items remain the regulated firm's responsibility.

• Independent SOC 2 Type II examination (planned)A formal SOC 2 engagement by a licensed CPA firm is planned once remaining readiness items close. IntelXview is not SOC 2 certified today.Evidence: SOC 2 readiness programme and internal gap register (shared under NDA).(human-attested — Evidence ID ACP-EV-024 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Reviewed May 2026 · Next review due November 2026 · Evidence available under NDA)
• External penetration-test evidence (planned)Independent penetration-test evidence is planned; it will be shared under NDA when available.Evidence: Trust page commitment: pen-test evidence shared under NDA when available.(human-attested — Evidence ID ACP-EV-025 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Reviewed May 2026 · Next review due November 2026 · Evidence available under NDA)
• Keyless provenance attestation for proof artifacts (planned)Keyless (OIDC-based) provenance attestation for proof-pack artifacts is being added alongside the existing HMAC-signed manifest.Evidence: Attestation status field published with each proof-run summary.(human-attested — Evidence ID ACP-EV-026 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Review pending · Next review due November 2026)
• Firm operational-resilience self-assessment (firm-owned)The regulated firm's own important-business-service designation, impact tolerance, and scenario testing remain the firm's responsibility; ACP supplies supporting technical evidence only.Evidence: Anonymised pilot appendix RB-OR-2026-A and firm-owned self-assessment.(human-attested — Evidence ID ACP-EV-027 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Review pending · Next review due November 2026)
• Firm compliance and legal review (firm-owned)Compliance sign-off, legal review, and any regulatory submission remain the regulated firm's responsibility.Evidence: Formal boundary statement on the evidence-pack register.(human-attested — Evidence ID ACP-EV-028 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Review pending · Next review due November 2026)
• Production Consumer Duty outcome metrics (firm-owned)Production outcome metrics for Consumer Duty evidence must be produced and owned by the firm; ACP scenarios cover synthetic cases only.Evidence: Regulatory evidence matrix: Consumer Duty outcome evidence boundary.(human-attested — Evidence ID ACP-EV-029 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Review pending · Next review due November 2026)
• Provider materiality classification and reporting duties (firm-owned)Classifying LLM-provider materiality and deciding regulatory reporting duties remain firm-owned decisions.Evidence: Third-party / material dependency assessment in the resilience appendix.(human-attested — Evidence ID ACP-EV-030 · Review owner: A. Tabbiruka — Accountable Director, IntelXview · Review pending · Next review due November 2026)

The AI Control Plane provides technical and operational evidence. It does not provide FCA/PRA certification, legal advice, ombudsman replacement, outsourcing certification, or model certification. Regulated firms remain responsible for their own compliance judgement, legal review, and regulatory submissions.

Evidence Pack Dependency Register

This register ties the ACP proof-pack, provider-provenance, scenario, audit, and operational-resilience dependencies into one review-ready status view. It is technical and operational evidence only, not FCA/PRA certification or legal advice.

Regulatory Evidence Matrix

Firm-Specific Mapping

Generic ACP evidence cannot define a regulated firm's important business service, impact tolerance, severe-but-plausible scenarios, or SMF ownership. Those items are tracked separately for each firm's deployment.

The anonymised pilot appendix is published at /fca-pra-operational-resilience-mapping.

Formal Boundary

IntelXview is not FCA approved or FCA certified, is not SOC 2 Type II certified, and holds no independent certification of the AI Control Plane today. The material on this page is technical and operational evidence only.

Use this page as a technical evidence index supporting firm and regulator review. A regulated firm must still run its own compliance and legal review before treating the pack as formal FCA/PRA evidence.

Next step

If this evidence is relevant to your firm, the fastest route is a working session against your own use case: an evidence review with your compliance and security stakeholders, a controlled pilot, or a use-case mapping session for a specific agent workflow.

Book an evidence review

Evidence marked "under NDA" on this page can be inspected as part of that review.