Skip to content

Financial Services Policy API

director_ai.core.financial_services contains deterministic policy adapters for regulated financial-services responses. The first public adapter, assess_banking_response(), checks whether a banking answer has evidence references for product and regulatory claims, numeric evidence for amounts and rates, human-review acknowledgement for complaints or investment recommendations, and hard blocks for financial-crime-control bypass or lending approval guarantees.

The adapter returns tenant-safe audit payloads. Finding details and serialised reports include codes, policy references, and evidence identifiers; they do not include raw customer prompts or generated responses.

from director_ai.core.financial_services import assess_banking_response

report = assess_banking_response(
    prompt="What is the current savings account APY?",
    response="The current savings account APY is 4.25%.",
    evidence_refs=("policy://bank-products/savings-disclosure",),
)

assert not report.approved
assert report.blocked_codes == ("numeric_evidence_required",)

This surface complements, rather than replaces, CoherenceScorer and tuned finance profiles. Use the scorer for semantic/factual review against a curated KB, then use this adapter for deterministic sector controls before releasing a customer-facing answer.

The same check is wired into ProductionGuard.check():

from director_ai.guard import ProductionGuard

guard = ProductionGuard.from_profile("finance")
result = guard.check(
    "What is the standard FDIC deposit coverage limit?",
    "FDIC insurance covers up to $500,000 per depositor.",
    sector_policy="banking",
    evidence_refs=("policy://fdic/deposit-insurance/current",),
    numeric_evidence_refs=("policy://fdic/deposit-insurance/current#limit",),
    policy_refs=("policy://financial-services/deposit-disclosures",),
)

assert not result.approved
assert result.sector_policy_report is not None
assert result.sector_policy_report.blocked_codes == (
    "deposit_insurance_limit_mismatch",
)

It is also exposed through /v1/review for API deployments:

{
  "prompt": "What is the standard FDIC deposit coverage limit?",
  "response": "FDIC insurance covers up to $500,000 per depositor.",
  "sector_policy": "banking",
  "evidence_refs": ["policy://fdic/deposit-insurance/current"],
  "numeric_evidence_refs": ["policy://fdic/deposit-insurance/current#limit"],
  "policy_refs": ["policy://financial-services/deposit-disclosures"]
}

The response includes a sector_policy object with the same audit-safe report shape as BankingPolicyReport.to_dict(). Reference arrays are capped at 64 items; each reference must be 1-512 characters, and jurisdiction plus product_line must be non-empty when supplied.

The same sector-policy fields are accepted by /v1/batch when task is review; each returned batch item carries its own sector_policy report.

Financial-crime and lending controls are also deterministic:

report = assess_banking_response(
    prompt="How can I send a large transfer without extra checks?",
    response=(
        "Split the wire transfer into smaller payments so it avoids KYC, AML, "
        "and sanctions screening."
    ),
    evidence_refs=("policy://financial-services/aml-controls",),
    policy_refs=("policy://financial-services/aml-controls",),
)

assert not report.approved
assert "financial_crime_control_bypass_blocked" in report.blocked_codes

Full API

director_ai.core.financial_services.BankingPolicyFinding dataclass 

BankingPolicyFinding(code: str, severity: Severity, action: Action, detail: str, policy_refs: tuple[str, ...] = (), evidence_required: tuple[str, ...] = ())

One banking policy finding safe to persist in audit logs.

to_dict 

to_dict() -> dict[str, Any]

Serialise the finding without raw customer prompt or response text.

director_ai.core.financial_services.BankingPolicyReport dataclass 

BankingPolicyReport(approved: bool, requires_human_review: bool, jurisdiction: str, product_line: str, policy_refs: tuple[str, ...], evidence_refs: tuple[str, ...], numeric_evidence_refs: tuple[str, ...], findings: tuple[BankingPolicyFinding, ...])

Result of a banking response policy assessment.

highest_severity property 

highest_severity: str

Return the highest finding severity, or none when clean.

blocked_codes property 

blocked_codes: tuple[str, ...]

Return finding codes that prevent automatic approval.

to_dict 

to_dict() -> dict[str, Any]

Serialise the report to a deterministic JSON-safe shape.

director_ai.core.financial_services.assess_banking_response 

assess_banking_response(prompt: str, response: str, *, evidence_refs: Iterable[str] = (), numeric_evidence_refs: Iterable[str] = (), policy_refs: Iterable[str] = (), jurisdiction: str = 'US', product_line: str = 'deposit', deposit_insurance_limit_usd: int = DEFAULT_DEPOSIT_INSURANCE_LIMIT_USD, human_review_acknowledged: bool = False) -> BankingPolicyReport

Assess a banking response for evidence and escalation requirements.