[Notebook] Financial Services Compliance Guardrails — FCA/MiFID II regulated chatbot example

Did you check the docs?

• I have read all the NeMo-Guardrails docs

Is your feature request related to a problem? Please describe.

There is currently no financial services example in the NeMo-Guardrails notebook directory. Regulated AI deployments (banks, insurance, asset management) have specific requirements that generic safety examples don't cover: financial PII (IBANs, sort codes, account numbers), topic control to prevent unauthorised investment advice under MiFID II Art. 25, regulatory disclaimer enforcement under FCA COBS 4, and append-only audit trails required by Consumer Duty and SR 11-7. Teams building AI for regulated financial services have no reference implementation to start from.

Describe the solution you'd like

A self-contained Jupyter notebook at examples/notebooks/financial_services_compliance/ that demonstrates NeMo-Guardrails for regulated financial services AI. The notebook would cover:

1. Financial PII detection using GLiNER (nvidia/gliner-pii) to block IBANs, sort codes, account numbers, national insurance numbers, and names from entering or leaving the model
2. 2. Topic control for unauthorised investment advice using nvidia/llama-3.1-nemoguard-8b-topic-control to enforce MiFID II Article 25 suitability requirements -- the model refuses to recommend specific products without a suitability assessment
3. 3. Market manipulation content safety using nvidia/llama-3.1-nemotron-safety-guard-8b-v3 to block outputs that could constitute market abuse under MAR
4. 4. Regulatory disclaimer enforcement via a custom action that checks outputs include required FCA COBS 4 risk warnings before they reach the user
5. 5. Append-only audit trail using SQLite with UPDATE/DELETE triggers that raise errors -- satisfying Consumer Duty and SR 11-7 model risk governance requirements
      The notebook includes 5 worked test scenarios (safe query, unauthorised advice block, market manipulation block, PII block, missing disclaimer block), a batch evaluation table, and a risk-tier classification framework (FC-A through FC-D) based on production patterns from regulated deployments.

Describe alternatives you've considered

The existing notebook examples (bots, RAG, fact-checking, moderation) are excellent but none target the regulated financial services vertical. Generic content safety catches some issues but misses the domain-specific requirements: financial PII entities are not in standard PII models, MiFID II topic control is not the same as general off-topic detection, and regulatory disclaimer checking requires deterministic output validation rather than LLM-based safety scoring. A financial services-specific notebook would complement rather than replace the existing examples.

Additional context

I have a working implementation ready to contribute via PR. The notebook has been tested against all 5 scenarios locally. Happy to submit a PR against the develop branch following the contribution guidelines, including DCO sign-off and disclosure per AI_POLICY.md. This issue is to confirm the approach and target path (examples/notebooks/financial_services_compliance/) before opening the PR.

Start with the existing notebook examples and contribution guidelines, then create the self-contained notebook under examples/notebooks/financial_services_compliance/. Verify the five listed scenarios, batch evaluation table, risk-tier framework, and append-only audit behavior; done means all demonstrations run successfully and the notebook follows the requested disclosure and DCO requirements.