Solutions · Banking and Insurance

Can you show the model's decision trail?

Your examiner does not want a policy document. They want to see the decision, the policy that ran, and the tamper-evident record that proves it happened the way you said it did. Raidu writes that record on every call, inside your own cloud.

Runs inside your AWS, GCP or Kubernetes install.

Last reviewed

Book a working session Watch it work

Illustrative. No customer data.

The difference

Their way, and ours.

Their wayRaidu
A system log that may or may not have captured it.A tamper-evident, chained record on every AI decision.
Evidence assembled the week before the audit.Evidence written as the decision happens.
Model policies in a document, not in the system.Policies that run, refuse, and record their decisions.

SR 11-7 was superseded on 17 April 2026. SR 26-2 leaves generative and agentic AI outside its model risk scope but expects banks' own governance to cover them. Raidu writes the evidence as each decision happens.

Source: Federal Reserve / OCC SR 26-2 model risk guidance, April 2026

What examiners ask

And the record that answers each one.

For Indian banks and NBFCs: what does the RBI expect on AI audit trails?

Only 18% of RBI-supervised entities that use AI maintained AI audit logs (RBI FREE-AI report, 13 August 2025). Raidu writes a write-once log entry for every AI interaction, inside your own infrastructure.

For Indian banks and NBFCs: what do the Digital Lending Guidelines require?

RBI’s Digital Lending Guidelines (2022) require data collection to be need-based, with prior explicit consent and an audit trail. Every AI decision and every consent is written to a tamper-evident record in your own cloud, ready to produce to an inspector.

For banks with US operations: where is the SR 26-2 model risk evidence?

Every call carries a tamper-evident record: which model ran, which policy checked it, what the model saw after personal data was replaced with coded stand-ins such as PERSON_1, and the verdict. SR 26-2 superseded SR 11-7 on 17 April 2026. SR 26-2 leaves generative and agentic AI outside its model risk scope but expects banks’ own governance to cover them. The record is the evidence for that governance.

Can customer data reach a third-party model?

No. Raidu intercepts the prompt before the model sees it. Personal data is replaced with coded stand-ins such as PERSON_1. The model receives the coded stand-ins, not the raw values. The replacement is logged and on record.

Raidu capabilities

The tools that build the trail.

AI Firewall

61 guardrail policies across 36 detection services, run on the full prompt before the model is called. Every decision on record.

Read more ›

Evidence and Audit

Write-once audit chain. Every record verifiable without calling back to Raidu.

Read more ›

AI FinOps

Spend by team, by model, by decision, with budgets and alerts.

Read more ›

The product

See the screens.

Product windows from the films, with sample data. No customer data.

AI Firewall ›

The record

The decision trail, on the record.

Every AI call becomes a tamper-evident entry: who sent it, what policy ran, what the model received, and what was decided.

Entry 6,217 Pre-send checkpoint · 10:44:22Z
Use case
Credit summary · risk classification
Personal data
Replaced with coded stand-ins
Policy
Model governance policy · passed
Verdict
Allowed and logged
Proof
Tamper-evident, on record
sha256 99d7a7d80c92a535b2342db3ac264a72…
entry 6,217 of 23,891 · write-once

The call path

Where the decision is made.

  1. InterceptEvery SDK and console call
  2. InspectPolicy checks on the full prompt
  3. DecideAllow or refuse before the model is called
  4. RecordTamper-evident record of the decision and policy
  5. ProveWrite-once chain entry, linked so any change shows

Also covered

  • Decision trail
  • SR 26-2 evidence
  • Personal data removed
  • Write-once audit log
  • India and US guidance

The clock

The regulatory calendar.

2 September 2022

RBI Digital Lending Guidelines: data collection must be need-based, with prior explicit consent and an audit trail for every lending interaction.

Reserve Bank of India, Digital Lending Guidelines 2022

13 August 2025

RBI FREE-AI report: only 18% of RBI-supervised entities that use AI maintained AI audit logs.

Reserve Bank of India, FREE-AI report

17 April 2026

SR 26-2 supersedes SR 11-7. Generative and agentic AI sit outside its model risk scope; banks' own governance must cover them.

Federal Reserve / OCC

Common questions

Answered straight.

What does SR 26-2 require?
SR 26-2 (Federal Reserve, 17 April 2026) replaces SR 11-7. Its principles apply to traditional statistical models and non-generative, non-agentic AI. Footnote 3 places generative and agentic AI outside its scope, while expecting banks’ governance practices to guide controls for those tools. Raidu keeps the record of each AI decision that such governance relies on.
How does Raidu protect customer data in AI calls?
Every prompt is intercepted before the model sees it. Personal data is replaced with coded stand-ins such as PERSON_1. The model receives the coded stand-ins, never the raw values. The substitution is logged and on record as part of the decision entry.
Can the audit records be verified independently?
Yes. Records are write-once and linked so any change shows. Your auditor can verify any record independently.
Where does the audit chain live?
Inside your own cloud account. Raidu deploys in your own cloud account.
Does the same framework cover insurance as well as banking?
Yes. The same policy stack, audit chain, and evidence trail applies to any financial institution that uses AI in decisions touching customers or risk models.

Related

Every AI decision, on a record your examiner accepts.

Book a working session