Kynara gives financial services firms per-action policy enforcement, human approval workflows, and a tamper-evident audit log that satisfies SOC 2, ISO 27001, PCI-DSS, GDPR, and GLBA requirements — before your AI agents touch customer data or execute transactions.
Financial firms face a hard constraint: regulators expect audit trails, access controls, and human oversight for any process touching customer accounts, transactions, or sensitive data — regardless of whether the actor is a human or an AI.
These patterns cover 80% of what financial services teams need when deploying AI agents in regulated workflows.
{
"effect": "require_approval",
"actions": ["payments.refund.issue"],
"condition": {
"op": "gte",
"args": ["ctx.resource.attrs.amount_cents", 100000]
}
}
hr_admin role condition.{
"effect": "deny",
"actions": ["crm.*"],
"condition": {
"op": "not_in",
"args": ["ctx.context.ip_country",
["US","GB","DE","FR","IE"]]
}
}
Financial regulators increasingly require demonstrable controls over automated decision-making systems. Here's how Kynara's features map to the frameworks your compliance team already uses.
| Requirement | Framework | Kynara feature | Status |
|---|---|---|---|
| Audit trail for all automated actions | SOC 2 CC6, ISO 27001 A.12.4 | SHA-256 hash-chained audit log, append-only trigger, CSV export | ✓ Built-in |
| Access control based on least privilege | SOC 2 CC6.3, NIST AI RMF | RBAC + ABAC with non-escalation guarantee | ✓ Built-in |
| Human oversight of high-risk decisions | EU AI Act Art. 14, NIST AI RMF | require_approval as first-class policy outcome | ✓ Built-in |
| Break-glass access with justification | SOC 2 CC6.7, PCI-DSS 7.2 | JIT grants with justification + ticket URL, full audit trail | ✓ Built-in |
| Anomaly detection and incident response | SOC 2 CC7, ISO 27001 A.16 | Z-score deny-rate alerting, geo-jump detection, auto-revocation | ✓ Built-in |
| Identity management and SSO | SOC 2 CC6.1, ISO 27001 A.9 | Okta, Azure AD, SAML 2.0, OIDC, SCIM provisioning | ✓ Built-in |
| Data residency and sovereignty | GDPR Art. 44–46 | Data residency middleware, regional endpoint enforcement, HTTP 451 | ✓ Built-in |
| GLBA Safeguards Rule — access controls for customer financial data | GLBA 16 CFR Part 314 | RBAC + ABAC with non-escalation guarantee; full audit log per access event | Built-in |
| PCI-DSS — access control to cardholder data environments | PCI-DSS Req 7 & 8 | RBAC roles, JIT grants with justification, audit trail of every CDE access | Built-in |
| Custom retention and eDiscovery export | FINRA 17a-4, SEC 17 CFR 240 | Custom retention + S3 nightly export (Enterprise) | Enterprise |
In financial services, the principle of least privilege is non-negotiable. Kynara enforces it mathematically at every decision point.
When an AI agent acts on behalf of a user — say, a wealth management assistant acting for a client-facing advisor — the agent's effective permissions are the intersection of the agent's own role grants and the user's role grants.
This means: even if you accidentally over-scope an agent, it cannot do anything the dispatching human wasn't already allowed to do. This property is hardwired into the Kynara engine — it cannot be bypassed by policy configuration.
Pass on_behalf_of_user_id in the decision request and the intersection is applied automatically.
POST /api/v1/decisions/check
{
"subject_type": "agent",
"subject_id": "<wealth-agent-uuid>",
"on_behalf_of_user_id": "<advisor-uuid>",
"action": "portfolio.write",
"resource": {
"type": "portfolio",
"id": "port_cust_8812",
"attrs": { "amount_cents": 500000 }
},
"context": {
"time": "10:30",
"ip_country": "US",
"mfa_verified": true
}
}
// Agent has portfolio.write AND advisor has portfolio.write
// → effective scope: portfolio.write ✓
// Agent has admin.override but advisor does not
// → effective scope: denied before policy eval
Free plan includes 3 seats, 10,000 decisions/month, full policy engine, audit log, and approval workflows. Enterprise plans include custom retention, dedicated deployment, and a dedicated compliance review.