Reported by Simon Daniel Yusuph l Journalist at Weng Global
Nigerian banks and financial technology companies have been urged to prevent artificial intelligence (AI) agents from independently authorising payments, amid growing concerns that poorly controlled automated systems could trigger widespread financial losses and disrupt the country’s interconnected digital payment infrastructure.
Raj Kamal, founder and chief executive officer of TransFi, a payments technology company, warned that allowing AI agents to execute financial transactions without adequate safeguards could expose banks, fintech companies, businesses and customers to significant risks.
His warning comes as financial institutions increasingly explore AI-powered services to improve payment processing, customer engagement, fraud detection and operational efficiency.
Kamal argued that although AI can help simplify financial services, institutions must ensure that automated systems cannot independently approve and settle transactions without appropriate controls.
The central concern is that an AI agent could misinterpret a customer’s instructions, respond to manipulated information or repeat a software error across multiple accounts. If the system has authority to execute payments, a single mistake could potentially affect numerous transactions before the problem is detected.
For Nigeria, where digital transfers have become an essential part of everyday commercial activity, the warning raises questions about how banks can adopt emerging technologies without undermining the security and reliability of payment services.
TransFi CEO Calls for Stronger Controls Over AI Payments
Kamal recommended that financial institutions initially restrict AI agents to interpreting customer instructions, processing information and recommending actions, while retaining payment authorisation and settlement within established control systems.
Under this approach, AI would assist with decision-making without automatically receiving unrestricted authority to move money.
The distinction is important because conventional software generally operates according to predefined instructions, while advanced AI agents can interpret requests, use connected digital tools and carry out multiple tasks with limited human intervention.
In financial services, these capabilities could allow an AI agent to receive a payment request, identify a recipient, prepare a transaction and initiate the process through connected banking systems.
However, the same capabilities could create additional risks if the agent misunderstands the request or follows malicious instructions embedded in emails, websites or other information sources.
In an interview with The PUNCH, Kamal referred to a three-layer approach in which AI interprets a customer’s intentions, while authorisation and settlement remain governed by predictable rules.
The approach would allow financial institutions to benefit from automation without making an AI model the final authority over customers’ money.
Kamal also advised banks and fintech companies to begin with narrowly defined applications where potential mistakes would be inexpensive and transactions could be reversed without extensive disruption.
Such gradual adoption would allow institutions to assess how AI systems behave under different conditions before giving them access to more sensitive financial operations.
The recommendation reflects a wider challenge facing the financial sector: determining how much independence automated systems should receive when their decisions can directly affect people’s savings, business transactions and access to essential funds.
Why Autonomous AI Agents Present a Different Financial Risk
The risks associated with AI agents extend beyond traditional cybersecurity threats.
A conventional cyberattack may involve criminals gaining unauthorised access to a banking system, stealing customer credentials or manipulating transaction records.
An AI agent, however, could have legitimate access to a system and still initiate an incorrect transaction.
Because the agent may be operating with valid permissions, existing security checks could fail to identify the problem if the transaction appears technically legitimate.
For example, an agent instructed to settle an invoice might interpret a fraudulent message as a legitimate change of payment details. If it can independently approve and execute the transfer, the money could be sent to the wrong account without the customer intending to authorise that particular transaction.
This is an illustrative scenario rather than a reported incident involving a Nigerian bank.
Another concern is the speed at which automated systems can operate.
A human employee might make an error affecting one transaction before recognising the mistake. An autonomous agent connected to several financial services could potentially repeat a similar error across numerous accounts in a short period.
The danger could increase if several institutions rely on the same AI models, software providers or technical infrastructure.
A common weakness affecting widely used technology could expose multiple organisations to similar failures, creating risks that extend beyond a single bank.
Financial institutions must therefore assess not only whether an AI agent can perform a task successfully, but also how it behaves when instructions are unclear, information is manipulated or a connected system malfunctions.
Testing must examine failure scenarios as carefully as normal operations.
Previous Payment Errors Highlight Nigeria’s Exposure
Nigeria’s existing digital payment infrastructure provides important context for Kamal’s warning.
The country relies heavily on interconnected banking systems and electronic transfer services to support payments between individuals, businesses, financial institutions and public-sector organisations.
While these systems have improved the speed and convenience of transactions, technical failures can also create difficulties when large volumes of money move between accounts.
According to figures cited by Kamal in his interview with The PUNCH, a logic error involving the Nigeria Inter-Bank Settlement System’s NIP platform in September 2024 reportedly resulted in N13.66 billion being credited to 176 accounts across 19 banks and fintech companies without corresponding debits from the originating accounts.
Kamal said the problem was detected three days later, by which time N4.19 billion remained unrecovered.
The figures were reported as part of his explanation of the potential consequences of software failures in interconnected payment systems.
The incident illustrates how a technical error can create financial obligations involving several institutions, making detection, reconciliation and recovery more complicated.
Kamal also cited another incident involving a Nigerian bank in July 2024, when 718 customers reportedly transferred N1.13 billion into accounts held across 45 banks and fintech companies following a system glitch.
The affected institution subsequently sought to recover the money through legal proceedings, according to the account provided in the report.
These incidents involved conventional technical failures, not necessarily AI agents. Their relevance lies in demonstrating how errors within a connected payment environment can affect customers and institutions beyond the system where the problem originated.
Kamal warned that introducing autonomous AI agents without adequate safeguards could create an additional layer of risk because automated systems might initiate or repeat transactions at high speed.
However, the scale and likelihood of any future AI-related loss would depend on the design of the system, its permissions, the controls surrounding transactions and how quickly a malfunction could be detected.
The incidents should therefore be understood as illustrations of existing infrastructure risks rather than evidence that AI agents have already caused comparable losses in Nigeria.
Interconnected Payment Systems Could Spread the Impact
Nigeria’s digital payment infrastructure depends on cooperation between banks, payment processors, fintech companies and other service providers.
When a customer transfers money, the transaction may involve several systems responsible for initiating the payment, verifying the relevant information, crediting the receiving account and reconciling obligations between institutions.
The arrangement allows customers to move money conveniently, but it also means that a problem originating within one institution can create complications elsewhere.
If an automated system generates a large volume of erroneous transfers, receiving institutions may need to identify suspicious transactions, restrict affected accounts and determine which funds should be returned.
These actions could delay legitimate payments while the institutions investigate the problem.
Businesses could experience difficulties confirming incoming payments, while individuals might face delays in receiving funds needed for household expenses or other obligations.
The consequences could be particularly serious for small businesses that depend on reliable electronic transfers to pay suppliers, receive customer payments and manage daily operating expenses.
Kamal also highlighted the potential mismatch between immediate customer-facing transfers and the settlement processes used between financial institutions.
If erroneous transactions accumulate faster than institutions can reconcile their obligations, the resulting uncertainty could complicate efforts to establish the extent of the loss and recover the money.
A widespread incident could therefore affect payment reliability even at institutions that were not responsible for the original error.
These are potential risks rather than predictions that such a disruption will occur. Their seriousness would depend on the volume of affected transactions, the safeguards already in place and the speed of the response.
Nevertheless, the interconnected nature of Nigeria’s payment ecosystem makes coordinated risk management important as financial institutions introduce more sophisticated automation.
CBN Urged to Establish AI Payment Safeguards
Kamal called on the Central Bank of Nigeria (CBN) to establish safeguards before autonomous AI systems become more deeply integrated into the country’s financial infrastructure.
Among his recommendations is a requirement for the proposed National Payment Stack to identify transactions initiated by AI agents.
Such identification could help financial institutions distinguish automated activity from conventional customer-initiated transactions, making it easier to monitor unusual patterns and investigate incidents.
He also recommended transaction-velocity limits that could temporarily pause suspicious activity without shutting down an entire payment system.
These controls would allow institutions to respond to unusual transaction volumes while limiting the disruption caused by broader system shutdowns.
Another proposal is the creation of a register of AI agents authorised to initiate payments.
A register could help establish which automated systems are permitted to interact with financial infrastructure and what level of authority they have been granted.
Kamal further advocated coordinated incident reporting through the Nigeria Inter-Bank Settlement System, alongside common standards for technology vendors and regularly tested emergency shutdown mechanisms.
Shared reporting could help financial institutions identify recurring weaknesses and learn from incidents affecting other organisations.
Common vendor standards could also improve consistency in the way banks assess the security, reliability and accountability of external AI technologies.
Emergency shutdown mechanisms would provide another layer of protection by allowing institutions to suspend an automated system when it behaves unexpectedly.
These recommendations reflect Kamal’s position on the need for stronger controls. They should not be interpreted as confirmation that the CBN has adopted or approved each proposal.
Their implementation would require appropriate regulatory consideration, technical planning and coordination among the organisations responsible for Nigeria’s payment infrastructure.
Human Oversight and Transaction Limits Remain Important
Beyond regulation, banks and fintech companies would need to establish internal policies governing how AI agents interact with customer accounts and payment systems.
One important measure would be to ensure that an AI agent cannot automatically increase its own transaction permissions or bypass existing approval procedures.
Financial institutions could also impose transaction limits, restrict the accounts an agent can access and require additional verification when a payment falls outside an established pattern.
For higher-risk transactions, a separate authorisation process could provide an opportunity to identify incorrect payment details before money is transferred.
Institutions would also need monitoring systems capable of detecting unusual transaction volumes, repeated payment attempts and activity inconsistent with a customer’s instructions.
The controls should be designed to contain errors quickly rather than relying exclusively on investigations after money has already moved.
Regular testing would also be necessary to determine whether an agent can be manipulated by misleading instructions or unexpected information from connected applications.
Smaller financial institutions may face particular difficulties implementing these measures because they have fewer resources available for specialised AI governance, cybersecurity and risk-management teams.
Kamal’s proposal for shared standards and coordinated incident reporting could help address some of these challenges by encouraging consistent practices across the industry.
However, safeguards must reflect the specific risks associated with each application. An AI tool used to answer customer questions does not necessarily require the same level of transaction control as an agent authorised to initiate transfers.
The objective should be to match the system’s permissions to the potential consequences of its actions.
AI Chatbots Are Not the Same as Autonomous Payment Agents
The growing use of AI in banking also makes it important to distinguish between different types of technology.
A customer-service chatbot may answer questions about banking products, explain transaction procedures or direct customers to the appropriate support channel.
Such a system does not necessarily have permission to move money or make decisions affecting customer accounts.
An autonomous AI agent can have broader capabilities, depending on how it is designed and connected to other systems.
It may be able to interpret a request, access relevant information, interact with external tools and execute a sequence of actions without requiring a person to approve every step.
The key issue is therefore not simply whether a bank uses artificial intelligence, but whether the technology has the authority to initiate or execute financial transactions.
A system that provides recommendations while leaving final approval to a separate control process presents a different risk from one that can independently initiate payments and interact with settlement infrastructure.
Financial institutions should communicate these distinctions clearly to customers and regulators.
Customers also need to understand what permissions they are granting when connecting external AI services to their accounts.
Providing access to financial information should not automatically mean granting an external agent unrestricted authority to transfer money.
Clear permission settings, transaction notifications and accessible procedures for withdrawing access could help reduce the risk of unintended activity.
Who Should Pay When an AI Agent Causes a Loss?
Another important question concerns responsibility when an AI agent makes an incorrect payment.
Potentially affected parties include the customer, the bank or fintech deploying the agent, the technology developer and other service providers involved in processing the transaction.
Determining responsibility could become complicated when a system operates as intended technically but produces an outcome that the customer never authorised.
Kamal argued that the institution deploying the AI agent and maintaining the customer relationship should take first-line responsibility for compensating affected customers.
He said customers should not have to determine which technology provider or software component caused a failure before receiving assistance.
Under the approach he advocates, a bank or fintech could compensate an affected customer and subsequently seek recovery from a vendor where contractual arrangements or applicable law support such a claim.
The proposal places emphasis on protecting customers from disputes between financial institutions and technology providers.
However, the allocation of liability would ultimately depend on applicable laws, regulatory requirements, contractual arrangements and the circumstances of an individual incident.
Technology vendors could also face questions about their responsibilities when defects, inadequate safeguards or misleading system behaviour contribute to a loss.
Kamal warned that some vendor contracts limit liability to fees paid over a specified period. Such limits could leave financial institutions exposed if a technology-related incident causes losses significantly greater than the contractual compensation available.
Insurance could help institutions manage some of these risks, but coverage would depend on the terms of individual policies.
Financial institutions would therefore need to examine whether existing insurance arrangements cover losses caused by their own AI systems acting autonomously, rather than assuming that conventional cyber insurance automatically provides protection.
Clear contractual terms, effective incident records and transparent customer compensation procedures could become increasingly important as autonomous systems gain wider access to financial services.
Nigeria’s Digital Payments Growth Raises the Stakes
The debate comes as electronic payments play an increasingly important role in Nigeria’s economy.
According to figures cited by Kamal, Nigerian institutions processed approximately 11 billion transfers worth N1.07 quadrillion in 2024.
The reported volume highlights the scale of activity taking place across the country’s digital payment ecosystem and the importance of maintaining reliable transaction controls.
For consumers, dependable digital payments affect everyday activities such as receiving salaries, paying bills and transferring money to family members.
For businesses, electronic transfers support sales, supplier payments, inventory purchases and other routine commercial operations.
A serious disruption could therefore affect more than the financial institutions directly involved in a technical failure.
At the same time, AI offers potential benefits to the sector, including faster processing of routine tasks, improved customer support and more efficient compliance operations.
The challenge for banks and regulators is to capture these benefits without allowing automation to weaken the systems that protect customers’ money.
Kamal’s position is that financial institutions should not abandon AI adoption, but should establish clear boundaries around what automated systems can do independently.
That approach would require institutions to evaluate AI systems not only on their speed and efficiency but also on their reliability, transparency and ability to operate safely when something goes wrong.
What Happens Next?
Nigerian banks and fintech companies face the task of determining how autonomous AI systems should be governed as they expand their use of automation.
Kamal urged financial institutions to use the period ahead of AI-related anti-money-laundering compliance deadlines he identified for 2027 and 2028 to establish appropriate safeguards, test failure scenarios and strengthen their oversight arrangements.
The specific requirements and applicability of those deadlines would need to be confirmed against relevant regulatory guidance.
For the CBN, the wider policy question is how to ensure that emerging technologies remain compatible with the security and reliability requirements of Nigeria’s financial system.
For banks and fintech companies, priorities include restricting AI permissions, establishing independent payment authorisation controls, monitoring unusual transactions and preparing effective emergency responses.
Technology providers will also need to demonstrate that their systems can operate within clearly defined limits and that weaknesses can be identified and addressed promptly.
Ultimately, the risks associated with autonomous AI payments will depend on how institutions design, deploy and supervise the technology.
As Nigeria’s digital financial services sector continues to develop, protecting customer funds will require more than adopting new technology. It will also require effective governance, clear accountability and safeguards capable of containing errors before they spread across the payment system.
Weng Global – stories beyond borders
Sources
- The PUNCH — Interview with TransFi founder and CEO Raj Kamal on the risks of unchecked AI agents in banking payment systems, published October 30, 2026.
- ITWeb Africa — Reporting on AI agents, payment fraud risks and liability questions involving financial institutions, published October 15, 2026.
