Liberty Bank Command Center
Financial Crime · Deep Fraud · Sanctions · Authenticity · Agent Security · Cyber Defence
Find risk faster. Understand why it matters. Keep an auditable record of how AI was used.
Evidence-led risk intelligence
Bring sanctions changes, exposure records, market/currency stress and source evidence into one controlled review. Assessments bind the exact input versions and evidence digest so later review can see what the analyst and AI worked from.
Deepfake & impersonation triage
Document, voice, video, identity, account-takeover and payment-diversion cases can combine provenance evidence with versioned detector signals. Detector output prioritises investigation; it never becomes an autonomous fraud verdict.
Provenance stays separate from suspicion
Authenticity evidence can be linked from the HORIZIN Authenticity Engine and other approved sources. Every case receives a recorded workflow disposition. Where evidence is insufficient to establish authenticity or fraud, the disposition is mandatory human adjudication / escalation rather than an unresolved UNKNOWN state or an automatic fake verdict.
No autonomous adverse bank action
Liberty can assess, correlate, preserve evidence and request a controlled human workflow. It does not autonomously block customers, reject payments, move funds, determine legal sanctions status or file regulatory reports.
Independent approvals
High and critical financial-crime/fraud proposals require two different eligible reviewers. The requester cannot approve their own proposal; the source case is version-and-digest rechecked and the approval window expires.
Agent Passport & Multi-Agent Watchtower
Short-lived agent identity, principal binding, permitted tools/data, revocation, evidence-backed trust and cross-agent monitoring for circular delegation, mutual approval and suspicious propagation.
Scoped security validation
Exact authorised targets and paths, request budgets, pacing, two-person scanner approvals, public-IP/SSRF safeguards, report-parser hardening and an organisation-wide emergency stop.
Encrypted evidence & signed audit
Encrypted evidence bodies, tamper-evident audit history, Ed25519-signed complete audit exports/checkpoints, independently pinned verification and encrypted backup/restore controls.
AI Software Supply Chain Passport
Track models, tools, APIs, MCP servers, runtimes, libraries and containers with versions, digests, SBOM references, vulnerability blockers and evidence-backed approval.
Enterprise SSO / MFA path
OIDC Authorization Code + S256 PKCE, signed ID-token validation, approved MFA-assurance checks, explicit subject-to-role mapping, secure session controls and immediate revocation.
Bank production gate
Production mode deliberately remains locked until an independently signed review covers the exact release, deployment, origin and required evidence categories—including penetration testing, identity, infrastructure, model/data validation, privacy and secure software supply chain.
v5.0.1 RC1
The current hardened build passed 242 automated tests in two complete non-overlapping test groups, plus Python/JavaScript checks. v5.0.1 strengthens request framing, trusted-proxy handling, password-verification resource controls, tenant-bound session checks, least-privilege data access, fail-closed restore behavior, signed release/configuration binding, detector provenance and agent-principal binding. Independent penetration testing, rebuilt dependency verification and bank production acceptance remain external gates.
Evaluate Liberty without exposing bank production.
A bank pilot can start with synthetic or bank-approved non-production data in a dedicated, expiring evaluation environment, then progress only if the evidence supports it.
Hardened and penetration-test ready—not “hack-proof”.
HORIZIN does not market any system as impossible to attack. Liberty is engineered to fail closed around identity, approvals, target scope and production acceptance, and to make external testing measurable. Production still requires the bank’s approved identity, network/WAF, KMS/HSM, database/HA, SIEM, privacy, sanctions/entity-resolution, model-validation, penetration-test and risk-owner controls.