Decisions
Blocked / Flagged
Your Review
Vulnerability
Index
Score
| Seq | Time | Initiator | Decision | Cluster | Codes Triggered | Excerpt | Hash |
|---|
dbc-hr-api has implemented autonomous safeguards across its database schema, middleware, and sovereign pre-processing engine to satisfy these obligations.
| Article | EU AI Act Requirement | DBC Platform Safeguard | Technical Components / Database | Status |
|---|---|---|---|---|
| Article 9 |
Risk Management System Establish continuous risk estimation, identification, and mitigation process. |
Stateless preprocessing sovereign engine checks logic against 150 HR & 150 safety rules. | src/governance/engine.ts |
ACTIVE (PASS) |
| Article 10 |
Data & Data Governance Ensure data minimization, privacy controls, and bias detection/mitigation. |
Bidirectional PII Gateway tokenizes names, emails, phones, IBANs, and credit cards on ingress. | src/governance/pii-gateway/Table: pii_events |
SANITISED (PASS) |
| Article 11 |
Technical Documentation Compile detailed technical specs and dynamic compliance verification. |
E2E compliance integration test harness runs continuous verification suite. | scripts/e2e-eu-ai-act.tsOutput: eu_ai_act_mapping.md |
VERIFIED (PASS) |
| Article 12 |
Record-Keeping Enable automatic, lifetime event logging and decision audit trails. |
Chained SHA-256 ledger block chain prevents log modification. Superadmin access blocked via forced RLS. | src/services/audit-store.tsTable: audit_ledger (RLS active) |
LEDGER SECURED (PASS) |
| Article 13 |
Transparency to Users Operate transparently to enable interpretation of decisions & limitations. |
System prompt injections automatically append flagged rule descriptions to downstream LLMs. | src/index.tsPayload: system_prompt_injection |
TRANSPARENT (PASS) |
| Article 14 |
Human Oversight Enable natural persons to oversee, review, and override AI decisions. |
Flagged queries are enqueued automatically into the append-only HITL review queue. | src/services/hitl-store.tsTable: hitl_queue |
ENQUEUED (PASS) |
| Article 15 |
Robustness & Cyber Security Achieve cybersecurity controls, resilient to prompt injections and jailbreaks. |
Adversarial rules block malicious jailbreak inputs. HMAC_SECRET check enforced on server boot. | src/api/middleware/auth.tsRate limiting middleware active |
SECURED (PASS) |
Article 9 (Risk Management):
The rules engine runs in pre-check mode, scoring every query's potential for bias and hallucination vulnerability before downstream API delivery. Thresholds automatically block high-risk queries.
Article 10 (Data Governance & Privacy):
The 3-stage PII Gateway enforces state-of-the-art privacy by sanitising prompt inputs. Ingress names, emails, addresses, phones, credit cards, and bank credentials are tokenized into session-vault tokens (e.g. [NAME_a1b2c3d4]). Plaintext sensitive data is never saved to the database. Raw values are safely restored on egress before client delivery.
Article 12 (Immutable Record-Keeping):
Every transaction is logged to audit_ledger. Every entry contains a SHA-256 cryptographically chained hash computed over current row data and the previous row's hash. Postgres Row-Level Security (RLS) is forced, making the ledger append-only and immune to update/delete operations even by superadmins.
Article 13 (Downstream Transparency):
When compliance guidelines (like culture-fit bias or wage equity rules) are triggered, the engine dynamically injects transparency payloads (system_prompt_injection). This payload forces the LLM to provide clear inline citations (e.g., 🛡️ AI-DBC-006) for its decisions.
Article 14 (Human Oversight & HITL):
Flagged decisions automatically register a record with a unique identifier in the Human-in-the-loop (HITL) review queue. Reviewers can approve, reject, or escalate items directly in the queue widget.
Article 15 (Cybersecurity & Robustness):
Adversarial prompt injection rules block malicious inputs from bypassing system rules, defending downstream integrations from jailbreaks and unauthorized alterations. Server boots fail immediately unless a valid HMAC_SECRET is configured, protecting credentials.
| Tenant | Plan | Domain | Calls (30d) | Keys | Status | Last Active | Actions |
|---|
| ID | Type | Severity | Status | Summary | Reporter | Submitted |
|---|---|---|---|---|---|---|
| Loading… | ||||||
| ID | Type | Severity | Status | Summary | Reporter | Action | Date |
|---|---|---|---|---|---|---|---|
| Loading… | |||||||