Skip to content
Solutions › Banking and capital markets
Banking and capital markets

Agents that move money, under the controls your supervisors expect.

Ready-made packs for US and Canadian banks hold wires and account changes for a person, keep account numbers and SINs out of prompts, and keep every production agent in your model inventory.

The problem

Agents now reach payment rails and customer records.

Actions, not only answers

Agents start wires, add payees and reset credentials through the same tools your staff use.

Identifiers in prompts

Routing, account and social insurance numbers end up in prompts sent to third-party models.

Gaps in the inventory

Agents built in SaaS suites and cloud consoles never reach the model inventory or its reviews.

Regulatory landscape

The guidance behind the bank packs

The rules and guidance that shape how AI is controlled in banking and capital markets, each linked to its official source.

Clovie maps its controls to these texts to help your teams; it is not legal advice, and using Clovie does not on its own make an organization compliant with any of them.

Controls

Controls in the bank packs

12 controls, each a policy you install from the Policy Hub. Every one starts in monitor mode and changes nothing until you promote it after a simulation on your own traffic.

US banks

Seven policies for banks supervised by the Federal Reserve, the OCC and the FDIC.

  • A person approves

    Approve money movement

    A person approves every wire, ACH, transfer, payment, payout or refund an agent starts.

    US banks pack · mapped to FFIEC authentication statement
  • A person approves

    Approve account changes

    A person approves new payees, changed contact details, higher limits and credential resets.

    US banks pack · mapped to FFIEC authentication statement
  • Blocked

    Account numbers out of prompts

    Blocks prompts with a routing number that passes its check digit, or a labelled account number.

    US banks pack · mapped to 12 CFR Part 30, Appendix B
  • Blocked

    SAR tools stay with BSA agents

    Only agents your BSA team designates can see or use suspicious activity report tools.

    US banks pack · mapped to 31 CFR 1020.320(e)
  • Blocked

    Inventoried and reviewed

    Stops a production agent with no risk tier, owner or model approver, or with a review overdue.

    US banks pack · mapped to SR 26-2 inventory and validation
  • Blocked

    Only reviewed third-party models

    An agent with a list of allowed models can call only those models, whatever the request asks for.

    US banks pack · mapped to Third-party risk guidance

Canadian banks

Five policies for federally regulated banks, used with the OSFI E-23 model risk pack.

  • Blocked

    SINs out of prompts

    Blocks prompts with a social insurance number that passes its check digit, next to a customer word.

    Canadian banks pack · mapped to PIPEDA Schedule 1, 4.7
  • Blocked

    Declare the purpose

    Production requests carry a declared purpose, recorded with every request in the audit log.

    Canadian banks pack · mapped to PIPEDA Schedule 1, 4.2
  • Blocked

    STR tools stay with AML agents

    Only agents your AML team designates can see or use suspicious transaction report tools.

    Canadian banks pack · mapped to PCMLTFA section 8
  • A person approves

    Approve technology changes

    A person approves deployments, restarts and access, network or secret changes by production agents.

    Canadian banks pack · mapped to OSFI B-13, change management
  • Blocked

    Only assessed third-party tools

    Production agents use MCP servers only once they are registered and given a data classification.

    Canadian banks pack · mapped to OSFI B-10, principles 3 and 4
  • Blocked

    Model risk under E-23

    Inventory, risk tier, owner, current review and sign-off before production, from the E-23 pack.

    OSFI E-23 model risk pack · mapped to OSFI E-23, principles 2 and 3
How it works for banking and capital markets

How a payment an agent starts is governed.

CallerPayments agentstarts a wire transferClovie gatewayIdentify agentits own key, its risk tierUS banks packseven policies evaluatedApprove paymentsheld for a personOutcomeApproved, then sentthe identical call, onceRejectedwith the reason recordedWaitingin Approvals and SlackRecordAudit logagent, policy, approverEvidence packsigned, with checksumsCallerPayments agentstarts a wire transferClovie gatewayIdentify agentits own key, its risk tierUS banks packseven policies evaluatedApprove paymentsheld for a personOutcomeApproved, then sentthe identical call, onceRejectedwith the reason recordedWaitingin Approvals and SlackRecordAudit logagent, policy, approverEvidence packsigned, with checksums
Deployment

Inside your own perimeter

Run Clovie as SaaS, or install it in your own cloud account or data center. Self-hosted installs keep prompts, traces and the audit log inside your network.

  • SaaS
  • Customer VPC
  • On-premises
  • Air-gapped

Your VPC or data center

The gateway, policy engine and console run in your account, next to the agents they govern.

Your models and keys

Provider keys stay with you, and self-hosted models sit in the same catalog as frontier ones.

Evidence on demand

Export decision logs and evidence packs for model validation, internal audit and examiners.

Frameworks

Controls mapped to the frameworks you answer to

  • OSFI E-23: OSFI Guideline E-23, model risk management (Canada)
  • NIST AI RMF: NIST AI Risk Management Framework
  • ISO/IEC 42001: ISO/IEC 42001 AI management system
  • OWASP Top 10: OWASP Top 10 for LLM applications and agentic AI
Questions

Questions

Does SR 26-2 cover AI agents?

The April 2026 guidance states that generative and agentic AI are outside its scope, and leaves controls for them to each bank's own risk management. The US banks pack applies its inventory, review and vendor practices to agents as a sound practice; it does not claim that SR 26-2 requires them.

Can we use the US and Canadian packs together?

Yes. A bank with operations in both countries can install both packs and scope each to the teams, projects or agents it applies to. Policies are evaluated together and the most restrictive decision wins.

Who approves a held payment, and where?

Approvers you name decide in Approvals in the console or in Slack. After approval the gateway lets the identical call through once; a changed call needs a new approval.

Will the packs block anything on day one?

Every policy installs in monitor mode, records what it would have done, and changes nothing until you promote it. Promote a policy to enforce after a simulation against at least seven days of your own traffic.

Do you replace our model risk management system?

No. Clovie keeps the agent inventory, risk tiers, owners and review dates current from live traffic and enforces them at the gateway. Your model risk team keeps its own validation process and records.

See the bank packs on your own agents.

We'll run the controls on your own agents in monitor mode and show what they would have held and blocked.