| Version | Supported |
|---|---|
| 1.0.x | ✅ Active |
| < 1.0 | ❌ Not supported |
AlertIQ operates exclusively on synthetic data.
All transaction records, account profiles, alert histories, and investigation artefacts used in this project are generated by a purpose-built simulator. No real customer data, banking records, personally identifiable information, or actual financial crime case material is present anywhere in this repository or any associated datasets.
The simulation models realistic AML typologies (structuring, layering, shell company activity, etc.) but does not represent any specific institution, individual, or real financial event.
The following security controls are intentionally enforced and must not be removed or bypassed:
1. Ground truth suppression (true_sar not exposed via API)
The true_sar column from the alerts dataset is never returned by any API endpoint. Analysts work without knowing the ground truth label, which mirrors real investigation practice and prevents feedback loops. This is enforced by explicit column exclusion in alert_store.get_alert() and verified by tests/serving/test_temporal_integrity.py::TestAPITemporalIntegrity::test_get_alert_never_exposes_true_sar.
2. Temporal evidence integrity
The transaction history endpoint (GET /alerts/{id}/transactions) enforces WHERE txn_date <= alert_date. Investigators can only see transactions that occurred before the alert was raised - matching real AML compliance requirements and preventing hindsight bias. Verified by tests/serving/test_temporal_integrity.py.
3. Analyst notes not forwarded externally Investigation notes entered by analysts are stored in the local SQLite database only. Notes are never forwarded to any external service, LLM API, or logging aggregator that could expose case content.
4. Deterministic explainability Risk signal explanations are computed from stored feature values using deterministic thresholds - no external API calls, no LLM inference, no network requests during signal computation.
If you discover a security vulnerability in AlertIQ, please:
- Do not open a public GitHub issue. Security issues should be disclosed privately.
- Email a description to rithwikm7@gmail.com with the subject line
[AlertIQ] Security Vulnerability. - Include:
- A description of the vulnerability
- Steps to reproduce
- Potential impact assessment
- Any suggested remediation
You will receive an acknowledgement within 48 hours. For confirmed vulnerabilities, a fix will be prioritised based on severity.
AlertIQ is a portfolio demonstration project. The following security controls that would be required in a production AML system are not implemented:
| Control | Status | Production path |
|---|---|---|
| Authentication | Not implemented | JWT/OAuth2 middleware |
| Authorisation | Not implemented | Role-based access control (analyst vs. officer) |
| Rate limiting | Not implemented | Middleware + API gateway |
| Input sanitisation | Partial | Full input validation on all endpoints |
| Secrets management | .env file only |
HashiCorp Vault or cloud KMS |
| TLS enforcement | Cloud Run handles this | Enforced at infrastructure level |
| Audit logging to SIEM | Not implemented | Structured log forwarding |
| Session management | Not implemented | Stateful session tokens |
These limitations are documented explicitly because transparency about what is and is not implemented is itself a security practice.