AI Evidence for Regulated Entities · Triodian

Regulated entities

Your supervisor wants AI overseen inside your providers. Can they show you anything?

Banks, insurers, superannuation and pension funds are being asked to oversee AI within appetite, including AI inside third parties and the platforms behind them. Most of the record that would show it does not exist yet. The AI Use Assessment finds what can be produced, provider by provider.

The chain

One chain, and the same two questions at each boundary.

The obligation reaches the agreements with your providers. What evidence of AI behaviour satisfies it is still undefined, and most providers have not been asked.

Accountable

Your entity

Holds the agreements, sets the board's appetite, and carries the accountability.

What can you require?
What can be produced?
→

Third party

Material service provider

Core banking, policy or fund administration; claims or credit decisioning; fraud and AML monitoring.

What can you require?
What can be produced?
→

Fourth party

Platform behind it

The technology, pricing or AI model the provider relies on.

Second boundary · your own staff's use of external AI

Copilots and assistants used by your own legal, claims, credit and compliance staff. You control your own end of this boundary.

Three halves

Answered separately, each half is defensible and none is usable.

We work with large law firms and board risk advisers in alliance with Triodian, or with your own. Each party is engaged and paid directly; no referral fees in either direction.

The board decides

Appetite, tolerances and triggers

In facilitated sessions, recorded as the board's own positions. Your board risk adviser can facilitate.

Counsel decides

What can be required

Reach under each agreement, what is requirement and what is expectation, accountability steps, and privilege.

Triodian tests

What can actually be produced

The register of where AI acts, the evidence obtainable at each boundary, and a finding per arrangement.

The seven questions

Put to every arrangement on the register.

Questions 1 to 4 are ones your third-party risk work already asks. Questions 5 to 7 are the AI-specific ones. Each jurisdiction page restates them in its supervisor's terms.

1Does your supervisor's third-party or outsourcing regime reach this arrangement?Is the party within your group, or outside it?
2Is the provider a direct service to you, or part of a chain supporting one?A material service provider, or a platform behind one
3What rights do you hold?Passive · significant influence · contractual · control
4Where is each party, and does location change what can be obtained?Enforceability, data access and audit rights; offshoring
5What authority has been delegated to the AI?Productivity · analytical · decision · transactional · agentic · actuating
6What could failure do?Immaterial · contract breach · financial loss · customer harm · critical-operation disruption
7What evidence can reasonably be required?None · disclosure · attestation · event notification · audit · continuous telemetry · deterministic control

Two questions sit under all seven: what can you require, and what can actually be produced? The determination at each step is yours and your counsel's.

Where it stops

Diagnostic, not a sales process.

No legal advice

Every legal determination is yours and your counsel's. Legal questions go to counsel in writing.

No product presented

The measure is your supervisor's expectations and your board's appetite, not any product.

Aggregate evidence only

No customer data, no individual decisions, no system access, no access to a provider's model.

"No reasonable reach" is a finding

Where you hold no rights, that is recorded honestly. The work can stop at the findings.

Regulatory position as at 26 October 2026. Not legal advice: every legal determination is yours and your counsel's.