Command dashboard
Live view of the production platform: fraud metrics, model health, sanctions activity and security posture. (Sample data in this static preview — the deployed agent produces this live.)
£0
Fraud loss prevented (30d)
14,208
Transactions scored (24h)
37
High-risk alerts (24h)
9
SARs generated (30d)
0.8%
False-positive rate
12 ms
Median scoring latency
Recent high-risk alerts
| Time (UTC) | Pattern | Risk | Outcome |
|---|---|---|---|
| 14:02 | £48,000 CNP cross-border → NG, new account (2d), round amount | CRITICAL 0.93 | SAR generated |
| 13:47 | 12 transfers × £9,800 across 5 accounts (structuring) | CRITICAL 0.91 | GNN ring flagged |
| 13:31 | PEP counterparty, high-risk MCC 7995 | HIGH 0.78 | Enhanced due diligence |
| 12:58 | Geo-impossible: LHR → DXB → LGW in 40 min | HIGH 0.74 | SCA step-up |
| 12:20 | Sanctions near-match (OFSI, alias variant) | CRITICAL 0.97 | Blocked + OFSI report |
Security posture — 26 independent defense layers
HSTS preloadmTLS + PQ certsAI IDS + honeypots
Adversarial-ML defenceHSM keys3-of-5 MPC signing
MFA + ZK authHomomorphic scoringRASPSide-channel hardening
Immutable audit (SHA3-512)Threat intelSLSA supply chainContract security
ZK privacyRBAC zero-trustGDPR engineMonitoring
OMEGA: prompt-injection shield
AI-agent firewall
Adversarial-ML shield
Canary exfil grid
Zero-trust gateway
Model attestation
AI red-team loop
MITRE threat fusion
Tier: OMEGA-POST-QUANTUM-AI-ADVERSARY-PROOF · deep score 1.0 · 75 attack categories mitigated
Transaction risk analyzer
Score a transaction against the platform's risk model. Demo mode: this static preview runs a simplified client-side heuristic (sample output) — the deployed agent uses the full ML ensemble (XGBoost + random forest, 14 features) with adversarial-input defence.
Result
—
Fraud probability —
—
Sanctions screening
Fuzzy entity matching across five lists: OFSI (UK Consolidated), UN, HMT, EU and OpenSanctions — catches transliteration variants, aliases and near-misses.
Try “Viktor Petrov” (sample hit) or any other name (sample clean).
Result
—
GNN network analysis
The graph neural network places every event in the account/entity graph and detects patterns that row-based rules mathematically cannot see: mule rings (cycles), structuring/smurfing, and star (processor) patterns.
Detected mule ring (sample investigation)
Detection summary
| Pattern | Found | Detail |
|---|---|---|
| Mule ring (cycle) | YES | 6 accounts, funds return to origin |
| Structuring / smurfing | YES | 12 × £9,800 sub-threshold |
| Star / processor | WATCH | A-311: 41 inbound, 41 outbound |
| Velocity anomaly | HIGH | 3× median account velocity |
Investigation support: 4 Neo4j Cypher templates ship with the platform for casework (cycle expansion, shared-counterparty, timeline, geographic hops).
Zero-knowledge proof verification
Groth16 proofs on BN128: prove a property of a fraud score without revealing the underlying transaction or customer data. Four circuits ship with verified proofs and on-chain Solidity verifiers.
Result
—
Post-quantum report signing
Every significant finding is digitally signed so it is non-repudiable and tamper-evident. Production uses Dilithium-5 (NIST FIPS 204) for signatures, Kyber-1024 (FIPS 203) for key exchange and SPHINCS+ (FIPS 205) for stateless contexts — protecting SARs and audit trails against “harvest now, decrypt later”.
This static page computes a real live SHA-256 digest in your browser as a stand-in; the deployed agent signs with Dilithium-5.
Result
—
26-layer defence-in-depth
No single compromise is a breach. Layers 1–18 protect the platform; layers 19–26 (the OMEGA tier) are built specifically against autonomous AI-agent adversaries.
| # | Layer | Tier | What it stops |
|---|---|---|---|
| 1 | Network edge hardening | Base | HSTS preload, CSP, cross-origin isolation, mTLS with PQ certs, dynamic rate limiting |
| 2 | AI intrusion detection + honeypots | Base | 19 jailbreak/injection signatures, behavioural traps, instant bans |
| 3 | Adversarial-ML defence | Base | Out-of-distribution rejection, perturbation detection, model poisoning guards |
| 4 | Post-quantum cryptography | Base | Kyber-1024 / Dilithium-5 / SPHINCS+ (FIPS 203/204/205) |
| 5 | HSM key management | Base | Keys never leave secure boundary; hardware-backed storage |
| 6 | 3-of-5 MPC threshold signing | Base | No single key holder can sign fraud reports alone |
| 7 | MFA + zero-knowledge auth | Base | PSD2 SCA step-up; credentials never exposed to the platform |
| 8 | Homomorphic scoring | Base | Fraud scoring over encrypted features (Paillier) — plaintext never assembled |
| 9 | Runtime self-protection (RASP) | Base | Anti-debugger (ptrace/Frida), core-dump extraction, LD_PRELOAD implants |
| 10 | Side-channel countermeasures | Base | Constant-time crypto, timing jitter, memory locking, power-emission hygiene |
| 11 | Immutable audit log | Base | SHA3-512 hash chain; any tamper breaks verification immediately |
| 12 | Threat-intel reputation | Base | TOR/VPN, datacenter IPs, sanctioned-country, risky MCC, phishing domains |
| 13 | Supply-chain attestation (SLSA L3) | Base | Dependency confusion, typosquatting, unsigned images |
| 14 | Smart-contract security | Base | RBAC, replay protection, ZKP verification on-chain, formal checks |
| 15 | ZK privacy layer | Base | Properties proven without data disclosure (GDPR data minimisation) |
| 16 | RBAC + zero-trust architecture | Base | FCA / BANK / MERCHANT / REGULATOR roles, least privilege everywhere |
| 17 | GDPR compliance engine | Base | Hash-only PII, DPIA support, access audit, breach-notification support |
| 18 | Monitoring & observability | Base | Prometheus + Grafana, 8 custom fraud metrics, health-checked services |
| 19 | Prompt-injection shield | OMEGA | 4-layer: Unicode/homoglyph folding, recursive base64/rot decoding, 30+ multilingual signatures, structural heuristics + active neutralisation |
| 20 | AI-agent attack firewall | OMEGA | Fingerprints LLM agents (ChatGPT/Claude/Gemini/agentic frameworks), recon-sweep detection, LLM payload classification, auto-BAN |
| 21 | Adversarial-ML shield | OMEGA | FGSM/PGD perturbation detection vs calibrated baseline, float quantisation, training-poisoning manifests, drift alarms |
| 22 | Canary token grid | OMEGA | HMAC-signed canaries; any canary seen outbound = proof of exfiltration (human or AI) |
| 23 | Zero-trust agent gateway | OMEGA | 300s single-use signed tickets, per-action least privilege, replay rejection, risk-based re-auth |
| 24 | Model & artifact attestation | OMEGA | SHA-256 binding of models/modules/code + tamper-evident hash-chained ledger |
| 25 | AI red-team auto-loop | OMEGA | Continuous in-process attacks (18-attack library) against the live defences; detection-rate reporting |
| 26 | Threat-intel fusion | OMEGA | MITRE ATT&CK + ATLAS mapping, campaign correlation (LOW→CRITICAL), AI-adversary flag |
UK regulatory compliance
Compliance is built into the data flow, not added as a spreadsheet.
FCA (SYSC 4.1 / PRU)
Explainable scoring, immutable time-ordered decision audit, SAR workflow, counterparty checks against the live FCA Register API.
PSD2 / SCA
MFA with step-up thresholds and zero-knowledge authentication for privileged actions.
DORA
ICT risk management: health-checked services, incident-visible alerting, 4-hour notification path via GOV.UK Notify.
GDPR
Hash-only PII, zero-knowledge proofs, homomorphic scoring, differential-privacy tooling — data minimisation by architecture.
Sanctions Act 2018 / OFSI
Live OFSI Consolidated List plus UN / HMT / EU / OpenSanctions with fuzzy matching; matches surfaced with evidence.
SARs — POCA 2002 s.330
Structured filing endpoint; high-risk cases auto-generate a notification payload for GOV.UK Notify.
Live UK government integrations
| Service | Endpoint | Used for |
|---|---|---|
| FCA Register | register.fca.org.uk | Authorisation verification |
| Companies House | api.company-information.service.gov.uk | Entity status, directors, disqualifications |
| HMRC | api.service.hmrc.gov.uk | VAT / MTD entity checks |
| GOV.UK Notify | api.notifications.service.gov.uk | SAR / incident dispatch |
| OFSI Consolidated List | ofsistorage.blob.core.windows.net | UK sanctions screening |
Why this platform
Nine capabilities that most fraud-detection products simply do not offer.
🔬
Post-quantum security from day oneKyber-1024, Dilithium-5, SPHINCS+ (NIST FIPS 203/204/205). SARs, customer data and audit trails stay safe for their full retention life — your quantum migration is already done.
🤖
AI-adversary-proof by design (OMEGA tier)The AI agent is treated as the attacker: fingerprinted, injection-blocked, model-input attacks detected, and defences continuously self-tested by an autonomous red team.
🕵️
Privacy by design — score fraud without seeing the dataProve a score to a regulator without revealing the customer: zero-knowledge proofs + homomorphic encryption + hash-only PII. Smaller breach surface, simpler GDPR accounting.
🏛️
UK regulatory nativeFCA / PSD2 / DORA / GDPR / NIS2 / OFSI built into the data flow, with live government API verification and automated regulator notification.
⚖️
Evidence a regulator — and a court — will acceptPost-quantum-signed reports, blockchain-anchored, immutable hash-chained audit. Non-repudiable, tamper-evident, in dispute order.
🕸️
Sees organised fraud, not just bad rowsGNN detection of mule rings, laundering cycles and smurfing — the highest-loss fraud that per-transaction rules mathematically cannot find.
🌍
Multi-list sanctions intelligence with fuzzy matchingOFSI / UN / HMT / EU / OpenSanctions — catches aliases and transliteration variants that cause the expensive sanctions breaches.
🧪
Proven, not promised13-phase automated master suite (115/116), 40/40 smart-contract tests, 4 independently verified ZKP circuits — re-runnable at any time.
📦
One deployable unitDocker Compose, 11 health-checked services, 12-step startup, Prometheus + Grafana observability.