Charge dispute started by an AI agent

How can a customer start a charge dispute with an AI assistant and have it posted to the bank's systems without the identity check or the signature happening in the chat?

Let the AI assistant start the process and let Callvu run it. The assistant calls Callvu over MCP with the customer's intent; Callvu's evaluators assess the request, the customer is identified on a secure, process-controlled screen outside the chat, the dispute is taken and signed there, and the case is posted to the bank's dispute system with the Reg E or Reg Z clock running from the notice. The assistant reports the outcome; it never touches the account, the identity data or the signature.

In scopeComplex: multi-party handoffsCallvu moves: write back

Talk to an expert Try it in the sandbox

Last updated November 6, 2026.

The problem today

The bank's assistant can talk about a dispute but cannot file one, or files one without knowing who it is talking to.

Today, without Callvu

A customer tells the bank's AI assistant that a charge is not theirs. The assistant explains the process and gives a phone number, which is where the customer started. Or, worse, an assistant with write access takes the dispute in the chat: the customer's identity rests on the chat session, the dispute statement is a message, and the transaction list was read into a model's context.

An AI assistant is a door, not a system of record. Identity, consent and signature belong on screens the process controls, where the regulation's evidence is produced; the chat is where the customer asked. The Reg E or Reg Z clock starts at the notice, which means the moment the customer told the assistant, and that moment has to be recorded by something that keeps records.

Whether the bank's assistant is Callvu AI or the bank's own, the question for compliance is the same: what can it do, under which rules, and where is the trace?

How it runs

From the requestto the system of record.

A customer tells an AI assistant (the bank's own, or Callvu AI) that a charge is wrong. The assistant calls Callvu over MCP with the intent to start a dispute. The time of the request is recorded as the notice.

Customer: States the problem to the assistant, then identifies themselves, picks the transaction, states the claim and signs on a secure screen.
AI agent: Starts the process over MCP and relays status; it holds no account data, performs no step, and never sees the identity data or the signature.
Dispute analyst: Investigates and decides the case in the Decision Desk, as for any dispute.
Callvu: Evaluates the request, identifies the customer on a secure screen, takes the dispute, posts it to the bank's systems and keeps the trace.

The assistant starts the process
The assistant calls Callvu's MCP tool for a charge dispute with what the customer said. The six evaluator agents assess the request (completion risk, policy, data trust, compliance evidence, audit readiness, exception patterns), and the Completion Orchestrator decides: execute, execute with safeguards, escalate to a person, or block. The notice time is recorded.
Identity on a secure screen
The customer receives a link to a process-controlled screen (or is handed off within the bank's app) and is authenticated under the bank's policy. Nothing about identity passes through the chat.
The dispute on the screen
The transaction is picked from the card's activity on the secure screen, the claim is classified, the dispute form is signed, and evidence is uploaded, exactly as in the card dispute process. The assistant sees only that the step is done.
Write back
The case is posted to the bank's dispute system with the notice time, the classification and the signed form, the Reg E or Reg Z clock running, and the card blocked where the claim is unauthorized use. The assistant tells the customer the reference and what happens next; the analyst takes the case in the Decision Desk.

Where it ends: The dispute is in the bank's dispute system with the notice time from the assistant's call and the identity, claim and signature from the secure screen. The Decision Trace shows why the AI-started request was allowed and under what safeguards; the Audit Trace shows every step, in the chat and out of it.

Runtime enforcement

What is enforcedwhile the work runs.

The regulations preloaded in Callvu Studio and the bank's own policies, applied as steps that run, not lines in a binder. A step that breaks a rule does not run, so a completed run was compliant.

Regulations

Reg E or Reg Z error resolution (12 CFR 1005.11 and 12 CFR 1026.13): The error resolution clock starts when the bank receives the notice, oral or written; a notice given to the bank's AI assistant is a notice to the bank, so the time of the assistant's call is recorded and the investigation, provisional credit and result notice run from it under the track the card type sets.
GLBA Safeguards and authentication (12 CFR part 364 appendix B): Customer identification and account data are protected under the bank's information security program; a chat session is not an authentication, and the account data is shown only on a screen the bank controls after authentication.
E-SIGN (15 USC 7001): The dispute form is signed electronically on a process-controlled screen with the consent and the record retention the Act requires; a message in a chat is not a signature.
Model risk and AI governance expectations (interagency model risk guidance, SR 11-7 and OCC 2011-12): An AI assistant that starts a regulated process is subject to the bank's model risk and third-party controls; the evaluator assessment and the Decision Trace are the evidence that each AI-started action was evaluated and allowed.

Company policies

An AI assistant may start the dispute process over MCP; it may not perform a step, read account data into its context, or capture identity, consent or a signature.
Every AI-started request is assessed by the six evaluators before it runs, and the Completion Orchestrator's decision is recorded.
The notice time is the time of the assistant's call to Callvu.
The dispute system is the system of record; the chat holds nothing but the conversation.

Controls

MCP tool with a defined scope: start a dispute, read status.
Evaluator assessment and orchestrator decision on every request.
Hand-off to a process-controlled screen for identity, claim and signature.
Notice time recorded at the call.
Card block on an unauthorized-use claim.
Status back to the assistant without account data.

Clocks and deadlines

Reg E or Reg Z error resolution clock: Runs from the assistant's call to Callvu, which is the notice.
Secure screen link: Time-limited; the customer resumes from the assistant or the app if it expires.

Documents and checks

The assistant's request and the evaluator assessment with the orchestrator's decision
The authentication result on the secure screen
The signed dispute form and evidence
The case posted to the dispute system with the notice time
The status returned to the assistant

In the Decision Trace

Why the AI-started request was allowed, with each evaluator's assessment and the orchestrator's decision.
The authentication applied outside the chat.
The classification and the clock set from the notice time.
What was posted to the dispute system.

In the Audit Trace

The assistant's call and its time.
The hand-off to the secure screen and each step on it.
The card block where applied.
The posting and the status returned.

The handoff

Between the front doorand the back office.

The front door

The AI assistant, the bank's own or Callvu AI, over MCP. The door starts the process; identity, consent and signature stay on process-controlled screens.

The back office

The dispute analyst sees a case like any other, with the notice time and the signed form, and never a chat transcript.

The Decision Desk

A request the Completion Orchestrator escalates (a trust or policy concern raised by an evaluator) goes to a person in the Decision Desk before anything runs.
The analyst decides the dispute itself in the Decision Desk, as in the card dispute process.

Proof

In scope.Proof at scale, not one process at a time.

The kind of process Callvu runs in production for customers in this industry. Ask for a reference on this exact process. The results below are the industry's published proof; each claim code links to its record on agents.callvu.com.

More than 1,000 processes are live across Callvu customers.

Across Callvu customers

Banca Transilvania automated more than 100 workflows with Callvu.

BNK-04

Bank Leumi digitized all loan application and servicing workflows within six months, and dozens more processes since.

BNK-01

Six evaluator agents assess every invocation (completion risk, compliance evidence, policy interpretation, data trust, audit readiness, exception patterns), and one deterministic orchestrator decides: execute, execute with safeguards, escalate to a human, or block.

ENF-02 · Self-asserted · verified 2026-11-02

Every run produces a Decision Trace (why each AI action was allowed) and every workflow execution an Audit Trace (the chronological governance record).

ENF-03 · Self-asserted · verified 2026-11-02

Questions

What buyers ask next,answered plainly.

Does the assistant see the customer's transactions?

No. The transaction is picked on a secure screen the bank controls, after authentication. The assistant receives a status and a reference.

Which assistant can start it?

Callvu AI or the bank's own, over MCP, under the same evaluation and enforcement. The six evaluators assess every request before it runs.

When does the Reg E clock start?

At the assistant's call to Callvu, which records the notice time. The signature on the secure screen does not restart it.

What does compliance get?

A Decision Trace showing why the AI-started request was allowed and under which safeguards, and an Audit Trace of every step, inside and outside the chat.

Related processes

Card disputes and chargebacks Debit card and EFT disputes Service requests completed digitally Try it in the sandbox All banking and credit unions processes

Bring this process, or the one that costs you most.

We build it against your policies and your systems, train your team and stay with you through launch. You are live in weeks.