It depends on what the customer still has. If they forgot a password but hold their registered phone, an AI agent can take them through the bank's reset flow at depth 3. If they lost the device that proves who they are, resetting means trusting a new one, where takeovers begin, so stronger checks and a named person come in.
Why is a reset the riskiest routine request in banking?
A reset hands over the keys. Whoever completes it can see the account, add payees and move money. Fraudsters know this, and many takeovers start with a call or chat saying "I've lost my phone and can't log in." The request is entirely ordinary. So is the story. The only defence is what the bank can check independently of what the person says.
That is why the agent never decides who the customer is from the conversation. It decides which reset route the bank's policy allows, based on what the customer can still prove.
What decides the route?
Three situations cover most requests.
Forgotten password, trusted device in hand. The customer can approve the reset on the registered phone or with a code sent to it. The bank's reset flow does the proving. The agent starts the flow and explains the steps. It never sees or sets the new password.
Locked after too many attempts. The account is locked for protection. If the customer can authenticate with the trusted device, the agent can lift the lock under the bank's rules. If the lock was applied by the fraud team, only the fraud team lifts it.
Trusted device lost or replaced. The customer must register a new device. This needs identity proof that does not depend on the old device, such as a document check with a live selfie, a code posted to the registered address, or a branch visit. Some of these the agent can start; the decision to trust the new device follows the bank's policy and, where signals look wrong, a person.
What may the agent do, and where must it stop?
| Situation | The agent | Where it stops |
|---|---|---|
| Forgotten password, registered device available | Triggers the reset flow, depth 3 | Seeing or setting the password |
| Customer-caused lock, device available | Lifts the lock after authentication, depth 3 | Locks applied by the fraud team |
| New device registration | Starts the identity process | Any failed or partial check |
| Change of registered phone number at the same time | Never alongside a reset | A person verifies separately |
| Reset requested by a third party | Records the request | A person checks recorded authority |
| After a new-device reset | Applies the policy's cooling-off on payees and limits | Lifting the cooling-off early |
A password reset alone is covered in whether an AI agent can reset a customer's password. This article is about the wider question of restoring access, including devices.
An example: a lost phone and a request to change number
Take a message to the bank's chat on a Friday night: "Lost my phone on holiday. New phone, new number. Need to get into my banking to pay a hotel bill. Please reset my access and update my number."
The agent recognises two requests that should never be completed together: registering a new device and changing the registered phone number. Either one alone has a safe route. Together they would cut the real customer out entirely. The agent offers the document-and-selfie check for the new device and says that the phone number change will be handled separately, by post or in branch. The person in the chat stops responding.
The agent freezes nothing, because the account itself shows no unusual activity, but it flags the attempt to the fraud team. On Monday the real customer, who still has their phone, sees the bank's message about a reset attempt and confirms it was not them.
What controls make it safe?
- The agent cannot handle secrets. It triggers flows; the customer completes them. No password or code passes through the agent.
- Device and contact changes are never combined. A reset with a new device and a change of phone or email in the same session is held.
- Cooling-off after new-device resets. New payees and higher payment limits are restricted for a period set by the bank.
- Narrow actions on the core system. The agent calls specific reset and lock-release operations, not general account access; see giving an AI agent safe access to your core banking system.
- Every attempt recorded. Successful and failed resets are filed with the route and the checks used.
After a reset that may have followed a takeover, review third-party access too: see whether an AI agent can access a customer's account through open banking. For how the bank's systems fit together, see what a core banking system is. The general limits are in AI agents in banking: what they may and may not do.
Questions people ask
Can the agent reset access for a customer who has no smartphone?
Yes, through the bank's other routes, such as a card reader, a posted code or a branch visit. The agent explains the options; it does not invent a weaker one.
Should the customer be told about every reset?
Yes. A notice to the registered contact details after any reset lets the real customer spot one they did not make.
Can the agent reset business banking access?
Business users are reset under the business's own administration rules. Often an administrator at the business resets other users, and the bank resets the administrator after stronger checks.