AI-Powered Scams Small Business Checklist: Signs and Dual Approval

AI-Powered Scams Small Business Checklist: Signs and Dual Approval

Checklist of phishing signs and fake payment requests, plus a dual confirmation routine your team should follow before releasing any payment.

•

Want to apply this in your company?

Book a call with Juan →

An AI-powered scam against a small business usually arrives as a message, an email or a phone call that imitates a supplier, a partner or an accountant, with fluent language and convincing context. The most effective defense does not depend on identifying which tool the scammer used. It depends on a fixed routine: any request to pay or to change bank details is confirmed through a channel independent of the one that delivered the request. This article lists the signs your team should watch for and walks through the dual confirmation step by step.

Key takeaways

  • AI scams do not require anything resembling superintelligence. Fluent text, an imitated voice or image, and some context taken from social media, websites or data leaks are enough.
  • A payment request that comes with urgency, secrecy or a changed bank account is the signal that most demands a pause.
  • Dual confirmation uses an independent channel that was known in advance, never the channel where the request arrived and never the number printed in that message.
  • The person who approves a payment should not be the person who received the request or the person who edits supplier records.
  • The rule must be written down, trained with the team and reviewed regularly, like any other operating process.

Why these scams became more convincing

The old advice to distrust messages with spelling mistakes used to work well. Today, text generation tools produce messages without errors, in a formal or casual tone depending on the target. They can reuse public information about the company, such as the name of the accountant, the brand of a supplier or the month-end closing calendar. In some cases, scammers use short audio clips to imitate the voice of a partner. For that reason, recognizing a voice or a writing style is no longer enough proof.

When the board hears the term superintelligence in news coverage and presentations, it helps to separate what exists today from what is hypothetical. The operational problem today is the generative models already available to anyone, not speculative systems. The superintelligence glossary for executives helps keep that distinction clear in internal conversations.

Signs of phishing and fake contact

No single sign proves a scam, but a combination of two or three should stop the process and trigger the confirmation routine. Your team should watch for:

  • A sender domain that looks like the real one, with a letter swapped, or a domain different from the one used on earlier invoices and contracts.
  • A shortened link or a button asking you to revalidate a registration, account, password or digital certificate.
  • An unexpected invoice or billing attachment, especially one asking you to enable macros or install an application.
  • A message from a new WhatsApp number claiming to be the partner, the manager or the accountant, supposedly switching phones.
  • Time pressure, such as "it must be done today" or "do not mention this to anyone."
  • A request to leave the official channel and continue in another app or over a phone call.
  • An audio or video message asking for something out of the ordinary. Voices and faces can be imitated, so verification must happen through a different channel.

Signs of a fake payment request

Financial fraud often appears as an update to something that already exists: the supplier's account has changed, the payment slip was reissued, the invoice has a new due date. The finance team should be suspicious when:

  • A known supplier's bank account, Pix transfer key or payee has changed without a formal registration update process.
  • The payment slip or invoice shows an amount, due date or payee that differs from the contract or the purchase order.
  • An "updated" bill arrives right after a price adjustment or close to month-end.
  • A payment request appears in the name of a director outside business hours or on a public holiday.
  • Proof of a previous payment arrives as a photo or screenshot instead of being checked against the bank statement.
  • A large amount has been split into smaller installments, just below the threshold that requires additional approval.

Dual confirmation routine, step by step

Dual confirmation works when it is short, written down and applied without exceptions by leadership. A practical sequence for the team:

  • Stop. No payment or change to bank details is processed in the same channel where the request arrived.
  • Record. The person who received the request notes the date, requester, channel, amount, reason and attachment in a spreadsheet or control system.
  • Confirm through an independent channel. Call the supplier's number already on file, or speak with a known contact in person. Never use the phone number or link provided in the message itself.
  • Compare the data. Amount, payee, account, due date and invoice number must match the contract, the purchase order and the current supplier record.
  • Separate the requester from the approver. A second person approves, and the approval is recorded with the date and the responsible name.
  • Apply the limit. Payments above the amount set by leadership require an additional sign-off from someone with the authority to approve them.
  • Treat an account change as a registration change. Updated bank details take effect only after confirmation through an independent channel and a formal update of the record. The next payment is checked against the new record.
  • If there is doubt or a payment has already gone out, contact the bank immediately, keep the evidence (messages, emails, attachments, receipts), inform leadership and assess whether the event must be logged as an incident.

Illustrative scenario: the accountant's bill

Illustrative scenario: a services company receives a message from an unknown number that appears to come from its accountant. It includes a photo of a signature and asks for payment of a tax slip due the next day. The finance assistant follows the routine. She does not pay, logs the request, calls the accountant at the number on file and receives confirmation that nothing was sent. The message becomes a record, leadership is informed, and the team checks whether other employees received the same request. The key point is not that the assistant recognized the scam on her own. The key point is that the routine meant she did not have to decide alone.

Putting the routine in place without stalling operations

The routine can be introduced using the method we apply in AI projects: map the current payment and registration flow, limit who can change bank details and approve amounts, pilot the rule on real workflows for a defined period with records kept, then hand it over to the team and decide whether the rule needs adjustments. This sequence avoids a common mistake: writing a polished policy that nobody follows on the day an urgent bill arrives.

If the company uses an AI agent on WhatsApp for customer service or collections, the rule must be explicit. The agent does not confirm payments, does not accept a change of bank account and does not handle exception requests. It records the request, tells the customer the request will be verified, and passes it to a person. A company policy for AI tools should also state which data may be pasted into external tools, since supplier and customer information moves easily into those environments. To decide what leadership can delegate and what must stay with people, the guide on what founders should decide themselves is a useful reference.

When an incident involves personal data of customers, employees or suppliers, Brazil's data protection law (LGPD) requires security incidents to be reported to the national data protection authority (ANPD) in cases with relevant risk or harm to the data subjects. The decision should go through the company's data protection officer, with legal support when needed. For guidance on governance and risk management in AI systems, the NIST AI Risk Management Framework describes useful principles, such as human oversight and documented decisions.

Frequently asked questions

Is AI-powered fraud only a problem for large companies? No. Small businesses are frequent targets because they often have informal supplier records, few people approving payments and mixed communication channels, such as a partner's personal WhatsApp. A simple dual confirmation routine reduces that exposure considerably without requiring a dedicated technical team.

Does dual confirmation delay supplier payments? It can add a few minutes, but the check fits into the routine when supplier records are current and each supplier's contact is registered. Recurring payments to known suppliers with no change in data tend to move quickly, and the longer confirmation is reserved for anything new or out of pattern.

Can an AI agent on WhatsApp help check payment requests? It can log the request, collect the basic information and pass it to a person along with the conversation history. It should not confirm payments, validate a change of bank account or decide exceptions, because those decisions require verification through an independent channel and human approval.

What should we do if the team has already paid a scam? Contact the bank immediately, since speed affects the chances of blocking the transfer. Then keep all the evidence, inform leadership, assess with the data protection officer whether the event is an incident that must be reported, and review the routine to understand why confirmation did not happen.

Sources

AI scamsSmall businessPhishingPayment controlsFraud prevention