Opening Your APIs to Fintech Partners? An 11-Step API Security Checklist for Banks
Most bank partner-API incidents start with traffic that looks legitimate. Here are 11 ordered steps, from inventory and FAPI-grade tokens to runtime detection and offboarding, with typical tools and effort for each.
Why Partner-Facing APIs Need Their Own Security Sequence
You've signed your first aggregator, embedded lender or banking-as-a-service partner, and the go-live date is set. The hard part of API security for banks at this stage isn't encryption. Every serious team already terminates TLS properly. The real danger is that partner traffic is authenticated, well-formed and expected. When something goes wrong, it usually starts with a valid token reaching data it was never meant to see, or with a partner's credentials leaking and quietly running at scale.
Order matters. Skip the inventory and you end up securing the APIs you think you have. Buy runtime detection before tokens are tied to specific partners and you get alerts you can't attribute to anyone. The checklist below runs in the sequence we'd recommend for a mid-sized bank or wealth manager opening a first wave of five to twenty partners.
Regulation shapes the baseline too. Several open banking and open finance regimes require or strongly point toward the FAPI OAuth 2.0 profile from the OpenID Foundation. Examples include the UK's Open Banking standards, Brazil's Open Finance and Australia's Consumer Data Right, with Gulf regimes following similar lines. The EU's DORA treats partner connections as ICT third-party risk. India's Account Aggregator framework and Singapore's MAS technology risk guidelines set their own expectations. Check which regime applies in each country you operate in, and confirm the current requirements with your compliance team. Where none applies, FAPI is still the most defensible bar to build to.
| Step | Typical tooling | Typical effort | Elapsed time |
|---|---|---|---|
| 1. Live API inventory | API discovery tools, gateway and traffic logs | 0.5 FTE security plus platform support | 3–6 weeks |
| 2. Data classification and ownership | API catalog, CMDB | Product owners, about 0.25 FTE coordination | 2–4 weeks |
| 3. Threat modeling per API family | OWASP API Security Top 10, workshops | 1–2 days per API family | Ongoing |
| 4. FAPI-grade authorization server | Ping Identity, Curity, Authlete, Keycloak | 1–2 IAM engineers | 2–4 months |
| 5. Sender-constrained tokens | mTLS or DPoP, PKI | IAM plus gateway engineer | 4–8 weeks |
| 6. Object-level authorization | Policy engine (such as OPA), service code | Each API-owning dev team | Ongoing |
| 7. Gateway hardening | Kong, Apigee, Azure API Management, AWS API Gateway | 1 platform engineer | 3–6 weeks |
| 8. Bot and abuse protection | Akamai, Cloudflare, Imperva, HUMAN, F5 | 0.25–0.5 FTE tuning | 4–8 weeks |
| 9. Runtime anomaly detection | Salt, Traceable, Cequence, Wallarm; SIEM | 0.5 FTE SOC analyst time | 6–10 weeks |
| 10. Continuous testing | 42Crunch, StackHawk, external pen testers | Pipeline work plus annual tests | Ongoing |
| 11. Partner offboarding | Client registry, consent store, runbook | Runbook owner | 2–3 weeks to build |
The effort and time figures are rough planning estimates for a bank with an existing gateway and IAM team. A firm starting from scratch should expect roughly double.
Steps 1–3: Find Every API, Classify It, Model the Threats
Step 1: Build an inventory from traffic, not from documentation
Your developer portal lists the APIs you mean to expose. Traffic shows the ones that actually exist: deprecated v1 endpoints still answering, mobile back-ends that were never registered, and test hosts with production data behind them. API discovery tools inspect mirrored traffic, gateway logs or cloud configuration to build that real list. Salt Security, Traceable, Akamai's API security offering, Cequence and Wallarm all do this, and most major cloud providers offer basic discovery as well.
Typical 2026 market ranges for a standalone API security platform run from roughly USD 50,000 to 300,000 or more per year. The main pricing drivers are API call volume and the number of environments. Pricing varies by contract and region, so confirm with vendors. One red flag: a tool that only discovers what passes through your gateway. Shadow APIs, by definition, often don't.
Step 2: Classify data and name an owner for every endpoint
For each API, record what data it returns, who owns it, which partners may call it and which regulation applies. Payment initiation, account balances, KYC documents and investment holdings each carry different exposure, and each may fall under a different legal basis: GDPR in Europe, India's DPDP Act, Brazil's LGPD, and bank secrecy rules nearly everywhere. If an endpoint has no named owner, it doesn't get exposed to partners. That one rule clears out a surprising amount of risk.
Step 3: Threat-model by API family
Run a short workshop for each family, such as accounts, payments or customer profile, using the OWASP API Security Top 10 as a prompt list. Spend most of the time on broken object-level authorization, broken function-level authorization and unrestricted access to sensitive business flows. All three are logic flaws, not infrastructure flaws, so a scanner alone won't find them.
Steps 4–6: FAPI-Grade Authentication and Authorization That Holds Up
Step 4: Stand up an authorization server that meets the FAPI OAuth 2.0 profile
FAPI 2.0 takes OAuth 2.0 and removes the risky choices. It requires pushed authorization requests (PAR), PKCE, strong client authentication through private_key_jwt or mTLS, and sender-constrained tokens. It also rules out the implicit flow and plain bearer tokens. If a regulator in your market mandates FAPI 1.0 Advanced instead, follow the regulator. Several regimes are moving toward 2.0, so pick a product that supports both.
Products with OpenID Foundation FAPI conformance results include Ping Identity, Curity, Authlete and the open-source Keycloak. Ask any vendor for its current certification results, not a marketing claim. Commercial licensing for a regulated deployment typically falls between USD 50,000 and 250,000 per year. Keycloak has no license fee but needs one to two engineers who know it well. For most banks without a strong in-house IAM team, a certified commercial product is the better fit, because conformance drift after upgrades is a real operating cost you'd otherwise carry yourself.
Step 5: Bind every token to the partner that holds it
A bearer token works for whoever has a copy. Sender-constrained tokens, bound to the client's certificate through mTLS or to a key through DPoP, are useless if stolen without the matching private key. Banks with mature PKI and partners that are themselves regulated institutions tend to choose mTLS. DPoP is easier for smaller fintech partners and mobile clients. Whichever you pick, configure the gateway to reject unbound tokens on partner routes. Don't simply log them.
Step 6: Check object-level permissions on every call
Here's a hypothetical. A partner holds a valid token scoped to accounts:read for Customer A. It changes the account ID in the path to one belonging to Customer B. If your service checks only the scope, it returns B's balance. The fix belongs in the service or in a central policy engine such as Open Policy Agent: on every request, confirm that the resource belongs to the consent attached to the token. Make this a code-review gate and a test case for every endpoint. It's ongoing work for each development team, so budget it into their capacity rather than handing it to security.
Steps 7–9: Gateway Controls, Bot Defense and Runtime Detection
Step 7: Treat API gateway security as policy, not plumbing
Kong, Google Apigee, Azure API Management, AWS API Gateway, MuleSoft and IBM API Connect can all enforce the controls that matter. Most banks underuse them. At minimum, partner routes should have:
- Strict schema validation against the published OpenAPI spec, rejecting unknown fields and oversized payloads
- Per-partner and per-consent rate limits and quotas, not only global limits
- Token validation at the edge, including audience, expiry and the binding from Step 5
- Response filtering so internal fields never leave the perimeter, even if a service over-returns
- Separate gateways or workspaces for partner, internal and mobile traffic
Typical 2026 enterprise gateway licensing runs from roughly USD 30,000 to 250,000 or more per year, depending on call volume and deployment model. Confirm local pricing with vendors.
Step 8: Add bot protection where APIs face the public or handle credentials
Partner-to-bank calls carry strong client authentication, so bots matter less there. The bigger exposure is everything around those calls: consent and login pages, onboarding APIs, card-validation endpoints and mobile back-ends. Credential stuffing, card testing and account-opening fraud run through those. Akamai, Cloudflare, Imperva, HUMAN and F5 offer behavioral and device-signal detection. Expect roughly USD 40,000 to 250,000 per year, plus a tuning period of several weeks so you don't block legitimate customers. Be wary of any product whose main control is CAPTCHA. It does nothing against headless API clients and frustrates real users.
Step 9: Baseline normal partner behavior and alert on drift
Runtime anomaly detection learns what each partner normally does and flags changes. Typical signals are a sudden sweep across sequential account IDs, a new geography, an endpoint the partner has never called, or a spike in 403 errors that looks like probing. The discovery platforms from Step 1 usually include this. Send their alerts, together with gateway and authorization-server logs, to your SIEM (Splunk, Microsoft Sentinel, Google SecOps or similar). Then write SOC playbooks that name the partner, the consent and the action to take. Plan for about half an analyst's time during the first quarter while you tune out noise.
Steps 10–11: Test Continuously, Then Offboard Partners Without Leftovers
Step 10: Put API tests in the pipeline and pen-test at least once a year
Static spec auditing (42Crunch is a common choice) catches weak definitions before code ships. Dynamic testing (StackHawk, Burp Suite Enterprise or your existing DAST tool) catches behavior at runtime. Neither reliably finds authorization logic flaws. That's what a specialist API penetration test is for. A focused test of a partner API family typically costs roughly USD 15,000 to 60,000 per engagement, depending on scope and region, and rates are often lower in India, parts of Southeast Asia and Latin America than in London or New York. Some regimes add specific testing requirements, such as threat-led penetration testing under the EU's DORA for in-scope firms. Check with your compliance team.
Step 11: Write the offboarding runbook before you need it
Partners get acquired, lose licenses, breach contracts or simply stop paying. When that happens, you need to cut access within hours, not weeks. The runbook should cover:
- Disable the client registration in the authorization server and revoke all refresh tokens
- Revoke or block the partner's certificates and signing keys
- Remove the partner's gateway routes, credentials and IP allow-lists
- Mark the partner's customer consents as revoked and notify affected customers where local rules require it
- Confirm data return or deletion under the contract, and record the evidence for auditors
- Run a final check against traffic logs that shows zero successful calls after the cutoff
Test the runbook with a dummy partner in staging before your first real partner goes live. Data deletion obligations and customer notification rules differ widely between jurisdictions, so have qualified legal counsel in each market review the consent and data-handling steps.
If budget forces a choice, fund Steps 1, 4, 5 and 6 first. A bank that knows every API it has, issues FAPI-grade sender-constrained tokens and checks object ownership on every call has closed the routes most partner API incidents use. Gateway tuning, bot defense and runtime detection then make that foundation visible and enforceable. They can't replace it.
Questions buyers ask
Do we need FAPI if our country has no open banking mandate?
It isn't legally required, but it's the most defensible standard available for partner access to account and payment data. FAPI 2.0 removes the OAuth configurations most likely to be exploited, and conformance-tested products already exist, so building to it costs little more than building a weaker custom profile. It also prepares you if a mandate arrives later.
Can our existing API gateway replace a dedicated API security platform?
For enforcement it can do a lot: schema validation, rate limits, token checks and response filtering. Gateways are generally weak at finding shadow APIs that bypass them and at learning each partner's normal behavior. Many banks start with gateway hardening and add a dedicated platform once they have more than a handful of partners or a large API estate.
Should we choose mTLS or DPoP for sender-constrained tokens?
mTLS fits when partners are regulated institutions with mature certificate management, or when your regime already uses certificates, as some open banking ecosystems do. DPoP is easier for smaller fintechs and mobile or browser-based clients because it doesn't need client certificates. Some banks support both and assign one per partner type.
How many people does this checklist realistically need?
For a mid-sized institution with an existing gateway and IAM function, a reasonable planning estimate is two to four people's worth of effort across security, IAM and platform engineering during the first six months. Development teams also take on object-level authorization work. Firms without an IAM team or gateway should plan for more, or use a managed service partner.
What is the most common mistake banks make with partner APIs?
Relying on token scopes alone and never checking whether the specific account or record requested belongs to the consent behind the token. That gap lets a legitimate partner, or an attacker holding its credentials, read other customers' data by changing an ID. Fix it in code and policy, and test every endpoint for it.
Written by the Apex Hub Pulse editorial team. Prices, features and rules change often - confirm details with the vendor or a licensed professional before you decide.