[FinServAssessments] Add 69 FinServ GenAI risk checks (FS-01 to FS-69)

[mehtadman87]

Feature request

Add 69 FinServ-specific GenAI security checks (64 standalone + 5 upstream
extensions) to the AIML Security Assessment, derived from the
AWS guide for Financial Services risk management of the use of Generative AI
(March 2026).

Why

Financial-services customers building GenAI workloads on AWS must demonstrate
controls aligned with SR 11-7, FFIEC CAT, NYDFS 500.06, PCI-DSS 12.3.2,
DORA Art.6, MAS TRM 9, ISO 27001 A.12, ECOA, and the OWASP LLM Top 10. The
AWS FinServ GenAI Risk Guide (March 2026) names specific AWS mitigations
across 15 risk categories that are not currently covered (or only partially
covered) by the existing 52 SM/BR/AC checks in this framework.

What

• 64 new standalone checks shipped in a new Lambda function
  finserv_assessments/ (FS-01 through FS-69, with 5 IDs contributed as
  upstream extensions instead).
• 5 upstream-extension edits to existing checks rather than duplicate
  them:
  • FS-17 → SM-07 (Model Monitor data quality)
  • FS-18 → SM-23 (Model drift detection)
  • FS-19 → SM-22 (Model Registry approval workflow)
  • FS-23 → BR-06 (Knowledge Base data-plane CloudTrail logging)
  • FS-64 → BR-04 (Guardrail trace logging)
• 6 material gap checks (FS-64 to FS-69) covering mitigations called out
  in the Guide but absent from FS-01..63 and the existing BR/SM/AC checks:
  guardrail trace logging, KB data-source S3 event notifications, AgentCore
  end-user identity propagation, agent financial transaction value
  thresholds, API Gateway request body size limits, and prompt input
  validation.

Risk categories covered (15 total)

Range	Category	PDF §
FS-01 to FS-06	Unbounded Consumption	§1.2.11
FS-07 to FS-11	Excessive Agency	§1.2.9
FS-12 to FS-16	Supply Chain Vulnerabilities	§1.2.12
FS-17 to FS-21	Training Data & Model Poisoning	§1.2.14
FS-22 to FS-26	Vector & Embedding Weaknesses	§1.2.15
FS-27 to FS-30	Non-Compliant Output	§1.2.1
FS-31 to FS-34	Misinformation	§1.2.3
FS-35 to FS-38	Abusive or Harmful Output	§1.2.4
FS-39 to FS-42	Biased Output	§1.2.5
FS-43 to FS-46	Sensitive Information Disclosure	§1.2.6
FS-47 to FS-50	Hallucination	§1.2.7
FS-51 to FS-54	Prompt Injection	§1.2.8
FS-55 to FS-58	Improper Output Handling	§1.2.13
FS-59 to FS-60	Off-Topic & Inappropriate Output	§1.2.2
FS-61 to FS-63	Out-of-Date Training Data	§1.2.10
FS-64 to FS-69	Material Gap Checks (cross-category)	—

Deliverables

• New Lambda functions/security/finserv_assessments/ (app.py, schema.py,
  requirements.txt).
• Updates to template.yaml, template-multi-account.yaml,
  statemachine/aiml_security_assessments.asl.json,
  deployment/1-aiml-security-member-roles.yaml,
  deployment/aiml-security-single-account.yaml, and
  functions/security/generate_consolidated_report/app.py.
• Extensions to the existing BR-04, BR-06, SM-07, SM-22, and SM-23 check
  implementations (non-breaking detection/remediation refinements).
• New reference docs under docs/: SECURITY_CHECKS_FINSERV_COMMON.md,
  SECURITY_CHECKS_FINSERV_PART1_INFRA_CONTROLS.md,
  SECURITY_CHECKS_FINSERV_PART2_GUARDRAILS_CONTENT_SAFETY.md,
  SECURITY_CHECKS_FINSERV_PART3_APP_LAYER_AND_GAPS.md, and an updated
  mappings CSV.

Implementation status

Implementation is complete in a personal fork. A PR will follow shortly
from mehtadman87/sample-aiml-security-assessment#feature/finserv-risk-checks
targeting this repo's main.

Acceptance criteria

• New Lambda wired into both single-account and multi-account SAM
  templates.
• New Step Functions branch added to the parallel-assessments state.
• IAM permissions scoped to the minimum required by the new checks.
• Consolidated HTML report reads the new
  finserv_security_report_{execution_id}.csv.
• All existing GitHub Actions checks pass (ruff, cfn-lint, pytest,
  semgrep, secrets scan).
• Local ASH v3 scan with severity threshold MEDIUM exits clean.
• Reviewed and approved by the team.

[mehtadman87]

Closing as a duplicate of #22. Both issues describe the same feature request — adding 69 FinServ GenAI risk checks (FS-01 to FS-69) derived from the AWS FinServ GenAI Risk Guide (March 2026). Issue #22 is the canonical tracking issue and is referenced by PR #23, which implements the full feature and is currently under review. No separate work is required here.